Commit Graph
454 Commits
Author SHA1 Message Date
Michael Suchacz 9cd38b5373 feat!: remove the Tasks API from coderd and codersdk (#28789)
Removes the Tasks API and task orchestration from coderd and codersdk:
the `/api/v2/tasks` handlers, task middleware and search filters, task
name generation, autobuild pause/resume logic, task notification
enqueues, telemetry, and the codersdk task types. Regenerates apidoc,
the API reference docs, and `typesGenerated.ts`.

**Breaking change:** the Tasks API is removed; requests to
`/api/v2/tasks` fail. Tasks was deprecated in v2.34.

**Stack context:** Coder Tasks removal stack (CODAGT-642): #28743
frontend → #28788 CLI → #28789 API → #28790 provisioner → #29084
notifications → #29085 build reasons → #29086 tables → #28750
permissions. The last four PRs replace the single data-drop migration
that used to live in #28750; each carries one concern and its own
migration number.

**Why:** with the UI (#28743) and CLI (#28788) gone, this deletes the
server surface. Database tables, RBAC, and notification templates stay
until the top of the stack so every layer builds and migrates
independently.

**What to scrutinize:**
- The `enable-ai-tasks` option is removed outright (review feedback):
configs still carrying `enableAITasks` fail to parse until the key is
deleted.
- provisionerdserver now writes `HasAITask: false` on template/build
flags; the column and these lines are dropped in #29085.
- `codersdk.ResourceTypeTask` stays so audit logs recorded before the
removal keep rendering.

**Validation:** `go build ./...`, `go vet ./...`, `go test
./codersdk/...`, site biome and tsc, CLI goldens regenerated, pre-commit
hooks.

**Size:** +242/-10,792 over 76 files; additions are regenerated
docs/goldens and test fixture updates.

> Created by Xum (AI agent) on behalf of @ibetitsmike.
2026-09-11 01:48:56 +00:00
david-fraley fef0763820 feat: report template builder build failure reasons in telemetry (#28586) 2026-09-10 08:21:29 -04:00
Paweł Banaszewski 8472a6884f chore!: remove env-based AI provider configuration (#29053)
> AI tools were used to create this PR

Removes deprecated AI provider configuration through environment
variables, CLI flags, and YAML.
Deletion includes indexed providers and legacy OpenAI, Anthropic, and
Bedrock options under both AI Gateway and AI Bridge branding, eg.
`CODER_AI_GATEWAY_PROVIDER_<N>_*` or `CODER_AI_GATEWAY_ANTHROPIC_KEY`.
Removes the associated startup-seeding code.
Updates tests and documentation.

**Breaking change:** Remove [obsolete AI provider CLI flags and YAML
keys](https://coder.com/docs/ai-coder/ai-gateway/providers#database-management-of-providers)
before upgrading, otherwise Coder will fail to start with an `unknown
flag` or `unknown option` error. Provider environment variables are
ignored and no longer seed the database. Existing database-stored
provider configuration is preserved.
2026-09-09 10:16:24 +02:00
68ca033d67 fix: deliver the rest of the invalid_request class to the client (#28736)
**TL;DR**

Fifth of the stack, split out of #28045. #28450 delivered the errors
raised *after* the callback is trusted and left the parameter failures
behind. This PR delivers those, and replaces the implicit ordering that
decided where an answer went with an explicit classification.

| PR | What it does |
|---|---|
| #28007 | Schema: `codes.scope` and `tokens.scope`, so a negotiated
scope has somewhere to live. |
| #28167 | `ScopesCover` compares an allowlist against a request by
permission coverage rather than by name. |
| #28178 | The authorize endpoint negotiates the scope against the app's
allowlist and persists it on the code. |
| #28179 | The consent page lists the negotiated permissions, and
`invalid_scope` reaches the client's callback. |
| #28450 | The same delivery for every other error raised once the
callback is trusted. |
| **#28736 (this)** | The parameter failures #28450 left behind, and the
classification that decides where each answer goes. |

- Stacked on #28450. Review that first.
- Nothing about what a token can do changes here. This is delivery, not
enforcement.

**Contract change**

A rejected parameter now arrives at the app's registered callback with
`error=invalid_request`, a description naming every failing field, and
the request's `state`. Both verbs.

| Verb | Was |
|---|---|
| GET | static "Invalid Query Parameters" page, 400 |
| POST | `WriteOAuth2Error`, 400 |

An integrator watching for either now reads the error from its own
callback instead.

**Two failures still answer on Coder**, because in neither case is the
callback trustworthy yet:

- A `redirect_uri` that does not parse, or does not exactly match the
registration. Redirecting to it would defeat the check that just
rejected it.
- A `client_id` sent more than once, or naming something other than the
app the callback was matched against. Coder cannot tell whose
registration it is about to redirect to.

An absent query `client_id` is **not** in that group. `httpmw` also
reads the POST form body and the §2.3.1 Basic credential, so an absent
query parameter still names a client and its failure is deliverable.

**Precedence change.** An app whose *registered* callback does not
parse, or uses a rejected scheme, now answers 500 even when the request
also carries parameter errors; that used to answer 400. Decided before
any parameter is read, so the others are never detected. Only reachable
for an app row that bypassed registration.
`DangerousCallbackSchemeOutranksParseFailure` pins it.

**Also in this PR**

- **Unrecognized parameters are ignored**, as §3.1 requires, rather than
rejected by `ErrorExcessParams`. An OIDC `nonce` or a vendor extension
no longer fails the request. Repeats of parameters the endpoint does
read are still rejected.
- **Every unsupported `response_type` gets one code.** Read as text
rather than through the SDK enum, so a value with no Go constant behind
it answers `unsupported_response_type` like `token` does, instead of
splitting on whether the SDK names it.
- **A malformed `resource` answers `invalid_target`** (RFC 8707 §2), but
only when nothing else failed. A client retrying on `invalid_target`
would otherwise resend a request still broken in a field it never heard
about.
- **`error_description` is bounded and readable.** Capped at 2048
characters and marked `(truncated)` before the log field or the
`Location` header is written. Entries read `field: reason` joined with
`; `, not the parser's debug shape, whose details contain commas and
cannot be split apart again.
- **A repeated `state` is charged to the client, not the callback.**
Reading `state` shares a function with the `redirect_uri` match, so it
used to fall into that carve-out and answer on Coder.

**Where in the OAuth Flow**

<details>
<summary>Diagram: the three-way dispatch, and what licenses a
redirect</summary>

```mermaid
flowchart TD
    NEW["newAuthorizeResponse<br/>parses the registered callback, checks its scheme,<br/>exact-matches any redirect_uri, reads state"]

    NEW -->|"registration unusable"| C500["500 on Coder<br/>server_error, value logged not echoed"]
    NEW --> PARSE["extractAuthorizeParams<br/>reads every parameter, collects all failures"]

    PARSE -->|"no failure"| OK["GET: consent page<br/>POST: authorization code"]
    PARSE -->|"failure"| KIND{"authorizeFailure.kind()"}

    KIND -->|"corrupt registration"| C500
    KIND -->|"redirect_uri or client_id at fault"| CODER["400 on Coder<br/>RFC 6749 4.1.2.1 carve-out"]
    KIND -->|"anything else"| CLIENT["302 to the callback<br/>invalid_request or invalid_target, with state"]
```

</details>

- Authorize endpoint only, both verbs. Token exchange, refresh, and
revocation are untouched, and `ErrorExcessParams` still guards the token
endpoint.
- `validatedCallbackURL` becomes `authorizeResponse`, built only by
`newAuthorizeResponse`, which runs the scheme check and the exact match
in that order. Holding one is what licenses a redirect, so the ordering
is structural rather than a convention each call site remembers.
- The scheme is checked on the *registered* URL, since `p.RedirectURL`
returns the client's URI on a mismatch and checking that would blame the
app for a request it never made.
- `state` moves onto the type, out of the parameter lists of
`withQuery`, `errorURL`, `codeURL`, and `redirectAuthorizeError`, so no
call site can emit a response the client cannot correlate.
- `extractAuthorizeParams` returns one `authorizeFailure` instead of two
trailing values, and both handlers dispatch on `kind()` rather than
re-deriving precedence from field checks.

**What it satisfies**

- RFC 6749 §4.1.2.1: a failure that is not a bad redirection URI or
client identifier is reported through the redirection URI. That is the
whole PR. Its two exceptions "MUST NOT" be redirected to, hence the
carve-outs.
- RFC 6749 §3.1: unrecognized parameters MUST be ignored, and no
parameter may appear more than once. Both now hold here.
- RFC 6749 Appendix A: `error_description` is confined to the permitted
set, on the decoded value.
- RFC 8707 §2: a `resource` that is not an absolute URI without a
fragment answers `invalid_target`.
- RFC 7636 §4.4.1: a malformed `code_challenge` is rejected at the
authorization request rather than deferred to token exchange, where the
error would point at the `code_verifier`.
- Not yet: CRF-26 (the `server_error` sites) and CRF-13 (fragment
delivery), both as in #28450.

**Docs and Swagger.** `docs/admin/integrations/oauth2-provider.md` gains
entries for the redirected parameter errors, `invalid_target`, and the
two failures that stay on Coder. Both authorize verbs document their 400
and 500 responses and stop advertising `response_type=token`, which
meant declaring the parameter as a `string`, since `Enums` appends to
what swaggo derives from the type.

---

Split out of #28045
([PLAT-479](https://linear.app/codercom/issue/PLAT-479)). Fifth of the
stack, stacked on #28450.
## Manual Tests

Verified by hand against a local dev deployment
(`v2.37.0-devel+1145d35a47`, dev Postgres), in addition to the automated
suite. The build string was checked first, because `develop.sh` builds
from whatever the tree held when it started and every result below would
otherwise be describing the wrong binary.

Two scenario groups need state the API cannot produce. The scope
allowlist is reachable only through dynamic client registration, which
was enabled for the run and disabled again at the end. A registered
callback that does not parse, or that uses a blocked scheme, is refused
at registration, so those rows were planted directly with `psql` on a
purpose-built client and restored afterwards.

**28 scenarios, all passing.** No correctness or security defect was
found in the
code this PR changes. Seven observations came out of the run: six are
consistency,
documentation or diagnosability points, and one is a small defect that
predates
this PR.

<details>
<summary><b>Scenario summary, all 28</b></summary>

| # | Scenario | Result |
|---|----------|--------|
| 1 | A well-formed request still succeeds on both verbs, and the
consent page renders | Pass |
| 2 | A rejected parameter now reaches the client's callback on `POST` |
Pass |
| 3 | The same on `GET`, and the consent page does not render first |
Pass |
| 4 | The description names every failing field, joined with `; ` | Pass
|
| 5 | A mismatched `redirect_uri` is answered on Coder with no
`Location` on either verb | Pass |
| 6 | An unparseable `redirect_uri` likewise, and the echoed value is
HTML escaped | Pass |
| 7 | A repeated `client_id` is answered on Coder although the callback
was valid | Pass |
| 8 | A single query `client_id` cannot disagree with the resolved app
over HTTP | Pass |
| 9 | A `client_id` supplied only in the `POST` form body is delivered,
not withheld | Pass |
| 10 | A `{UPPERCASE}` `client_id` is delivered, and also succeeds end
to end | Pass |
| 11 | A repeated `state` is delivered rather than charged to the
redirect carve-out | Pass |
| 12 | A registered callback with a blocked scheme answers 500 on both
verbs | Pass |
| 13 | An unparseable registered callback answers 500, indistinguishably
to the caller | Pass |
| 14 | Corrupt registration outranks a deliverable failure and both
carve-outs | Pass |
| 15 | The corrupt value is logged with the app ID and never appears in
a response | Pass |
| 16 | Restoring the row restores normal behaviour with no restart |
Pass |
| 17 | Every unsupported `response_type` gets one code; an empty one is
`invalid_request` | Pass |
| 18 | An unsupported `response_type` is not recast as a missing
`code_challenge` | Pass |
| 19 | Unrecognized parameters are ignored, including a case-variant
`REDIRECT_URI` | Pass |
| 20 | A misspelled `redirect_url` is ignored and cannot smuggle a
destination | Pass |
| 21 | Repeated known parameters are still rejected | Pass |
| 22 | `resource` answers `invalid_target` alone and `invalid_request`
in company | Pass |
| 23 | A fragment in `resource` is rejected | Pass, with one gap |
| 24 | A long description is capped at 2048 and marked `(truncated)`, on
both verbs | Pass |
| 25 | The description is sanitized to the RFC 6749 §4.1.2.1 character
set | Pass |
| 26 | The registered callback query is retained except the reserved
names | Pass |
| 27 | Scope rejections still reach the callback, and no consent page
renders | Pass |
| 28 | Declining consent carries `access_denied` and issues no code |
Pass |

</details>

<details>
<summary><b>Observations, all 7</b></summary>

| # | Observation | Severity |
|---|-------------|----------|
| 1 | `invalid_request` is returned in three description shapes: the
parser's `Invalid query params: field: reason; ...` aggregate, the PKCE
validator's single message, and `scopeFailureResponse`'s. A client
cannot parse all three with one rule. | Consistency, reviewer call |
| 2 | The phase 2 runbook for #28045 asserts the old `field: x detail:
y` message shape, which this PR replaces. That doc needs a one-line
update. | Docs follow-up |
| 3 | `clientIDInDoubt`'s `default` branch is unreachable over HTTP,
because `httpmw` derives `app.ID` from the same query value the parser
reads. Defensive rather than dead, but the comment does not say so. |
Comment clarity |
| 4 | A `POST` supplying `client_id` only in the form body always fails
with `client_id ... is required and cannot be empty`, naming a parameter
it did supply. `httpmw` reads the body, `RequiredNotEmpty` reads the
query. Pre-existing. | Misleading diagnostic |
| 5 | A misspelled parameter is invisible from every angle at default
verbosity. The client gets a 302 with a code, its redirect is silently
replaced by the registered one, and the `ignoring unrecognized
authorization parameters` line is `logger.Debug` so it is not emitted. |
Diagnosability |
| 6 | `resource` sent twice answers `invalid_target` rather than
`invalid_request`. RFC 8707 §2 frames `invalid_target` as being about
the resource value; a duplicated parameter is a malformed request under
RFC 6749 §3.1. | Low, error code choice |
| 7 | `resource=https://api.example.com/#` (trailing `#`, empty
fragment) is accepted and stored with the `#` intact, so the persisted
audience is not textually equal to the fragment-free form. `url.Parse`
maps empty and absent fragments both to `Fragment == ""`. Pre-existing,
in `tokens.go`. | Defect, low severity |

</details>

**One note for anyone re-running this.** Every request below depends on
`$CHALLENGE` from the most recent `new_pkce` call. Forgetting to call it
sends an empty `code_challenge`, which fails with the fixed "is required
and cannot be empty" message at 98 characters and looks exactly like a
cap or a validator failing to fire. That cost one wrong measurement
during this run before the numbers below were taken.

<details>
<summary><b>Shell helpers used throughout</b></summary>

```bash
export BASE_URL=http://localhost:3000
export AUTH_HEADER="Coder-Session-Token: $(cat ./.coderv2/session)"
export PGPASSWORD=$(cat ./.coderv2/postgres/password)
export PGPORT=$(cat ./.coderv2/postgres/port)

pg() { psql -h localhost -p "$PGPORT" -U coder -d coder -tAc "$1"; }
plant_callback() { pg "UPDATE oauth2_provider_apps SET callback_url = '$2' WHERE id = '$1';"; }
urlenc() { jq -rn --arg v "$1" '$v|@uri'; }

# 43 unreserved characters every time. Base64url of the raw 32 bytes.
new_pkce() {
  VERIFIER=$(openssl rand 32 | base64 | tr -d '\n=' | tr '+/' '-_')
  CHALLENGE=$(printf '%s' "$VERIFIER" | openssl dgst -sha256 -binary | base64 | tr -d "=" | tr '+/' '-_')
  STATE=$(openssl rand -hex 16)
}

authz_url() {
  local url="$BASE_URL/oauth2/authorize?client_id=$1&response_type=code"
  url="$url&redirect_uri=$(urlenc "$2")&state=$STATE"
  url="$url&code_challenge=$CHALLENGE&code_challenge_method=S256"
  printf '%s' "$url"
}

# The core assertion of the whole run: what answer, delivered where.
answer() { curl -s -o /dev/null -D - -X "$1" "$2" -H "$AUTH_HEADER" | grep -iE '^HTTP/|^location:'; }

show_callback() {
  python3 - "$1" <<'PY'
import sys, urllib.parse
q = urllib.parse.urlparse(sys.argv[1]).query
for k, v in urllib.parse.parse_qsl(q, keep_blank_values=True):
    print(f"{k} = {v}")
PY
}
```

Fixtures, all registered through DCR:

```text
plat479-plain    http://localhost:9876/callback
plat479-query    http://localhost:9876/callback?tenant=acme&error=stale
plat479-corrupt  http://localhost:9876/callback   (overwritten by psql in 12 to 16)
plat479-reserved http://localhost:9876/callback?tenant=acme&code=FAKECODE&error=stale&error_description=oldmsg&state=oldstate&keep=yes
plat479-ci       http://localhost:9876/callback   (scope allowlist coder:workspaces.access)
```

</details>

<details>
<summary><b>1. Baseline: a well-formed request still
succeeds</b></summary>

```bash
new_pkce; answer GET  "$(authz_url "$APP_ID" http://localhost:9876/callback)"
new_pkce; answer POST "$(authz_url "$APP_ID" http://localhost:9876/callback)"
```

```text
=== GET ===
HTTP/1.1 200 OK

=== POST ===
HTTP/1.1 302 Found
Location: http://localhost:9876/callback?code=coder_dzvaCPf9W1_...&state=e321bd5d5b594a1fae8a2d5da654b84a
```

`GET` carries no `Location` at all, so the consent page really rendered
rather than redirecting. The `state` echoed is the one sent. The code
persisted as `coder:all`, this app having no allowlist, with
`redirect_uri` recorded because the request supplied one.

Also driven through a real browser rather than curl, so the consent
form's `nosurf` token is the one submitted rather than the session-token
header alone. Clicking **Allow** delivered `code` and `state` to the
callback. Both request shapes agree on the success case.

Not to be misread: `oauth2_provider_app_codes` holds one row after both
POSTs, not two. `ProcessAuthorize` opens its transaction with
`DeleteOAuth2ProviderAppCodesByAppAndUserID`, so a row count is not a
count of successful authorizations. That predates this PR.

</details>

<details>
<summary><b>2 to 4. A rejected parameter now reaches the client's
callback</b></summary>

```bash
new_pkce
BAD="$BASE_URL/oauth2/authorize?client_id=$APP_ID&response_type=code"
BAD="$BAD&redirect_uri=$(urlenc http://localhost:9876/callback)&state=$STATE"
BAD="$BAD&code_challenge=short&code_challenge_method=S256"
answer POST "$BAD"
answer GET  "$BAD"
```

```text
HTTP/1.1 302 Found
Location: http://localhost:9876/callback?error=invalid_request&error_description=Invalid+query+params%3A+code_challenge%3A+must+be+43+to+128+characters+from+the+unreserved+character+set+%5BA-Za-z0-9-._~%5D&state=e89f4d98d4c8c347e5ae25175748b731
```

Identical on both verbs. Decoded:

```text
error = invalid_request
error_description = Invalid query params: code_challenge: must be 43 to 128 characters from the unreserved character set [A-Za-z0-9-._~]
state = e89f4d98d4c8c347e5ae25175748b731
```

The `GET` body confirms no page was rendered, against the baseline:

| Request | Body size | Occurrences of "Allow" |
|---|---|---|
| Malformed `code_challenge` | 263 bytes | 0 |
| Well formed (scenario 1) | 4928 bytes | 2 |

263 bytes is Go's redirect stub. There is no consent form in it, so
there is no Allow button to press for a request the server has already
refused.

Multiple failures are reported together (scenario 4):

```text
error_description = Invalid query params: code_challenge: must be 43 to 128 characters from the unreserved character set [A-Za-z0-9-._~]; resource: must be an absolute URI without fragment
```

The separator is `; `, and no entry uses the older `field: x detail: y`
shape. The code is `invalid_request` rather than `invalid_target` even
though `resource` is one of the two failures, which is the scenario 22
guard firing.

The same request also carried `code_challenge_method=plain`, which is
invalid and is *not* mentioned, because the method is validated after
`extractAuthorizeParams` returns and a parse failure short-circuits
first. Sent alone it is rejected as `code_challenge_method 'plain' is
not supported; use 'S256'`, with no `Invalid query params: ` prefix.
That difference is observation 1.

</details>

<details>
<summary><b>5 and 6. The redirect_uri carve-out keeps the answer
here</b></summary>

```bash
new_pkce
EVIL="$BASE_URL/oauth2/authorize?client_id=$APP_ID&response_type=code"
EVIL="$EVIL&redirect_uri=$(urlenc http://evil.example/steal)&state=$STATE"
EVIL="$EVIL&code_challenge=$CHALLENGE&code_challenge_method=S256"
answer POST "$EVIL"; answer GET "$EVIL"
```

```text
HTTP/1.1 400 Bad Request
HTTP/1.1 400 Bad Request
```

**The absence in that output is the assertion**: no `Location` on either
verb, so nothing redirects to `evil.example`. Stronger, `evil.example`
appears zero times in either full response, headers and body together,
so the attacker-supplied host is not reflected into the page either.

```json
{"error":"invalid_request","error_description":"Invalid query params: redirect_uri: Query param \"redirect_uri\" must exactly match http://localhost:9876/callback"}
```

The `GET` page names the *registered* callback as the value the
parameter had to match, which is the app's own configuration and safe to
display.

An unparseable `redirect_uri` reaches the same outcome by a different
route, having never parsed at all:

```json
{"error":"invalid_request","error_description":"Invalid query params: redirect_uri: Query param \"redirect_uri\" must be a valid url: parse \"http://%zz\": invalid URL escape \"%zz\""}
```

That message echoes the caller's value, unlike the mismatch case, so the
`GET` path was probed with markup in it:

```text
input:  http://%zz"><img src=x onerror=BAD>
page:   ...must be a valid url: parse &#34;http://%zz\&#34;&gt;&lt;img src=x onerror=BAD&gt;&#34;...
literal `<img` in body: 0     entity encoded `&lt;img`: 1
```

Escaped correctly by the template layer. Recorded as a negative result
so a future change to the error page has something to regress against.

One note for the docs: this detail contains colons, so a client
splitting `field: reason` entries must split each on its **first** colon
only.

</details>

<details>
<summary><b>7 to 10. The client_id carve-out, and its
negatives</b></summary>

A repeated `client_id` in an otherwise perfect request:

```text
HTTP/1.1 400 Bad Request   (both verbs, no Location)
{"error":"invalid_request","error_description":"Invalid query params: client_id: Query param \"client_id\" provided more than once, found 2 times. Only provide 1 instance of this query param."}
```

This is the one carve-out not implied by the response's shape. The
`redirect_uri` matched the registration, so `canRedirect()` was true and
there *was* somewhere to send the answer; it was withheld because
`extractAuthorizeParams` assigns `failure.redirect` only when
`clientIDInDoubt` is false.

**The `default` branch of that function is unreachable over HTTP**
(observation 3). `httpmw` resolves the app from
`r.URL.Query().Get("client_id")`, so `app.ID` and the parsed value
derive from the same string and cannot disagree. Three experiments:

```text
one query client_id, bad challenge          -> 302 delivered
two DIFFERENT client_id values              -> 400, caught by the repeated branch
query names app A, form body names app B    -> 302 to app A; the body is ignored
```

The negatives matter as much as the positives. An identifier absent from
the query but present in the `POST` form body is **delivered**,
correctly:

```text
HTTP/1.1 302 Found
Location: ...error=invalid_request&error_description=Invalid+query+params%3A+client_id%3A+...is+required+and+cannot+be+empty%3B+code_challenge%3A+...
```

That is observation 4: the description says `client_id` is required for
a request that supplied it. With everything else valid the same shape
still fails, while the identical request carrying `client_id` in the
query succeeds with a code. `RequiredNotEmpty` reads the query; `httpmw`
reads the body. Pre-existing, and arguably the parser is the layer in
the right, since RFC 6749 §4.1.1 puts authorization parameters in the
query string and §3.2.1 is the endpoint that uses a form body.

A braced upper-case UUID is delivered, and succeeds end to end:

```text
{UPPERCASE} client_id, bad challenge -> 302, description names ONLY code_challenge
{UPPERCASE} client_id, all valid     -> 302 with a real code
GET, all valid                       -> 200, consent page, heading resolves to the app's real name
```

`uuid.Parse` accepts that spelling and `httpmw` resolved through it, so
a string comparison here would withhold an answer the client is entitled
to, for no reason it could diagnose.

</details>

<details>
<summary><b>11. A repeated state is delivered, not
withheld</b></summary>

```text
HTTP/1.1 302 Found   (both verbs)
error = invalid_request
error_description = Invalid query params: state: Query param 'state' provided more than once, found 2 times. Only provide 1 instance of this query param.
```

Reading `state` happens inside `newAuthorizeResponse`, the same function
that runs the `redirect_uri` match. A count of errors across those two
lines would charge this to the redirect carve-out and answer 400. It
answered 302, so the field-specific test is doing the work.

The asymmetry with scenario 7 is deliberate: a repeated `client_id`
leaves the server unsure whose callback it holds, a repeated `state`
leaves the callback fully settled.

The response carries **no `state`**, since `parseSingle` collapsed the
repeat to the empty string and `withQuery` only sets it when non-empty.
Unavoidable rather than wrong: picking one of two values could satisfy a
CSRF check the client meant to be strict.

</details>

<details>
<summary><b>12 to 16. A corrupt registration outranks
everything</b></summary>

DCR refuses both planted values, which is why `psql` is needed and why
500 is the right answer: a stored one is not something a client did.

```json
{"error":"invalid_client_metadata","error_description":"invalid redirect_uris: redirect URI at index 0: redirect URI uses dangerous scheme javascript which is not allowed"}
{"error":"invalid_client_metadata","error_description":"invalid redirect_uris: redirect URI at index 0 is not a valid URL: parse \"http://%zz\": invalid URL escape \"%zz\""}
```

```bash
plant_callback "$APP_CORRUPT_ID" 'javascript:alert(1)'
new_pkce
CORRUPT="$BASE_URL/oauth2/authorize?client_id=$APP_CORRUPT_ID&response_type=code&state=$STATE"
CORRUPT="$CORRUPT&code_challenge=$CHALLENGE&code_challenge_method=S256"
answer POST "$CORRUPT"; answer GET "$CORRUPT"
```

```text
HTTP/1.1 500 Internal Server Error
{"error":"server_error","error_description":"The application's registered callback URL is not usable"}
GET: HTTP/1.1 500, page reads "500 - Invalid Callback URL", 3346 bytes, 0 Allow buttons
```

The unparseable callback is **indistinguishable to the caller**: same
description, and a `GET` page of exactly 3346 bytes, matching byte for
byte. Two different defects in stored state, one answer. Correct, since
neither is actionable by the client.

Note the request carried no `redirect_uri` of its own and still failed,
because the scheme is checked on the *registered* URL before the
client's value is consulted.

Precedence was tested against all three competitors, not one:

| Also wrong in the request | Would answer alone | Actually answered |
|---|---|---|
| `code_challenge=short` | 302 to the callback | **500** |
| `client_id` repeated | 400 kept here | **500** |
| `redirect_uri` mismatched | 400 kept here | **500** |

Reading the source, this is stronger than precedence.
`newAuthorizeResponse` runs at `authorize.go:250`, before any of
`client_id`, `code_challenge`, `scope` or `resource` is read, and the
early return discards whatever `p.Errors` already held. In rows two and
three those failures are never detected at all, so `kind()` ranking
`failureCorruptRegistration` first is belt and braces for a failure
constructed some other way.

**The value is logged and never echoed.** A distinctive marker was
planted and five requests fired:

```text
[erro] coderd: oauth2 app has an unusable registered callback URL
  request_id=01ca4796-c743-4923-a06b-afc1f3decc95
  app_id=12f9ca3b-7a2c-4c54-8ed7-723698c62bd1
  callback_url="javascript:alert(\"plat479-6d-marker\")"
  error= redirect URI uses dangerous scheme javascript which is not allowed:
         codersdk/oauth2_validation.go:116
```

Five requests, five lines, in request order. Occurrences of the marker
in the POST body, the rendered GET page, and the response headers: **0,
0, 0**.

The `request_id` joins that line to the request log entry whose
`response_body` field holds the vague message, so the server's own
record confirms what it sent for the same request rather than a replay.
The log is also where the two causes diverge, naming `validateScheme`
and its source line, which the response cannot do.

Restoring the row restores normal behaviour immediately, with no
restart:

```text
POST -> 302 with a code      GET -> 200 consent page
```

That is a control rather than housekeeping: it rules out the app having
been poisoned lastingly, and rules out a cached parse outliving the
`UPDATE`.

</details>

<details>
<summary><b>17 and 18. Every unsupported response_type gets one
code</b></summary>

| `response_type` sent | `error` | `error_description` |
|---|---|---|
| `token` | `unsupported_response_type` | Only response_type=code is
supported |
| `banana` | `unsupported_response_type` | Only response_type=code is
supported |
| `code token` | `unsupported_response_type` | Only response_type=code
is supported |
| `CODE` | `unsupported_response_type` | Only response_type=code is
supported |
| `code_extra` | `unsupported_response_type` | Only response_type=code
is supported |
| empty string | `invalid_request` | Invalid query params:
response_type: Query param 'response_type' is required and cannot be
empty |

`token` has a Go constant behind it and `banana` does not, so their
agreement is what reading the value as text bought. Three rows are more
interesting than `banana`: `code token` is a legal RFC 6749 §3.1.1
space-delimited list used by OIDC's hybrid flow, `CODE` confirms the
comparison is case sensitive, and `code_extra` confirms it is equality
rather than a prefix match.

The empty string is correctly the exception. A valueless parameter is
*missing*, not unsupported, and RFC 6749 §3.1 requires it be treated as
omitted.

**The error is in the query, not the fragment.** `response_type=token`
is the implicit grant's own value, so a fragment would be arguable, but
this deployment advertises `"response_types_supported":["code"]` alone,
and a fragment is never sent to the server, so it would be unreadable to
the client's backend.

PKCE is not recast, which the controls show:

| `response_type`, no `code_challenge` sent | `error` |
|---|---|
| `token` | `unsupported_response_type`, zero mentions of
`code_challenge` |
| `banana` | `unsupported_response_type`, zero mentions of
`code_challenge` |
| `code` | `invalid_request`, naming `code_challenge` |

Without the `if params.responseType == responseTypeCode` guard, a client
sending `token` would be told its `code_challenge` was missing, add one,
resend, and be told the same thing again.

</details>

<details>
<summary><b>19 to 21. Unrecognized parameters ignored, repeats still
rejected</b></summary>

Eight unknown parameters, each on an otherwise valid request, every one
issuing a code:

| Unknown parameter | Result |
|---|---|
| `nonce`, `prompt`, `login_hint`, `acr_values`, `max_age`, `ui_locales`
| code issued |
| `code_challenge_methods=S256` | code issued |
| `REDIRECT_URI=http://evil.example/steal` | code issued, **to the
registered callback** |

Five of those are OpenID Connect Core parameters, so an OIDC client
pointed here degrades to plain OAuth2 rather than failing. The ignored
parameters are **dropped, not forwarded**: the callback receives only
`code` and `state`.

The last two rows are the ones with teeth. Query parameter names are
case sensitive, so `REDIRECT_URI` must be ignored, and a misspelled
`redirect_url` likewise:

```text
case variant alongside a valid redirect_uri : Location host localhost:9876
case variant as the only redirect parameter : Location host localhost:9876
redirect_url typo, no valid redirect_uri    : Location host localhost:9876, code issued
evil.example occurrences in any Location    : 0
```

The consent page renders for the typo case too, and even its cancel link
targets the registered callback. A parser matching names
case-insensitively would have redirected to `evil.example` with a valid
code attached.

**Observation 5.** The `ignoring unrecognized authorization parameters`
line was never emitted. Three requests carrying a distinctive marker
produced no new server output at all, because the call is `logger.Debug`
(`authorize.go:307`) and the deployment runs with `verbose: null`. The
behaviour is correct and required by RFC 6749 §3.1, but at default
verbosity the misspelling is invisible from every angle: the client sees
a 302 with a code, its redirect is silently replaced, and nothing is
logged. The comment claiming the typo "surfaces here" is optimistic.

Repeats of known parameters are still rejected, which the replacement of
`ErrorExcessParams` had to leave alone:

| Repeated parameter | HTTP | `error` | Answered at |
|---|---|---|---|
| `code_challenge` | 302 | `invalid_request`, two entries one field |
client callback |
| `code_challenge_method` | 302 | `invalid_request` | client callback |
| `response_type` | 302 | `invalid_request` | client callback |
| `redirect_uri` | 400 | withheld | **Coder** |
| `scope` | 302 | `invalid_request` | client callback |
| `resource` | 302 | `invalid_target` | client callback |

`redirect_uri` being withheld confirms the carve-out keys on the field
name rather than the kind of mistake. `code_challenge` collects two
entries for one mistake, because `parseSingle` collapses the value to
empty and the PKCE block then reports it missing. The `resource` row is
observation 6.

</details>

<details>
<summary><b>22 and 23. resource, its own code, and the fragment
rule</b></summary>

Both sides of the rule, since a check enforced too broadly would reject
a valid URN:

| `resource` sent | `error` | Stored `resource_uri` |
|---|---|---|
| `not a uri` | `invalid_target` | not stored |
| `/api` relative | `invalid_target` | not stored |
| `https://api.example.com` | none, code issued | as sent |
| `https://api.example.com/v1?q=1` | none, code issued | as sent, query
is not a fragment |
| `urn:example:resource` | none, code issued | as sent |
| empty string | none, code issued | `NULL`, same as omitted |

`invalid_target` applies only when `resource` is the sole failure. Add
any second failure in any other field and the code becomes
`invalid_request` naming both, which is the retry-loop guard:

```text
resource + bad code_challenge -> invalid_request, both named
resource + repeated scope     -> invalid_request, both named
resource + repeated state     -> invalid_request, both named
resource alone (control)      -> invalid_target
```

Fragments are rejected in all three placements, with or without a path:

```text
https://api.example.com/#x         -> invalid_target
https://api.example.com#x          -> invalid_target
https://api.example.com/v1#frag    -> invalid_target
https://api.example.com/#          -> code issued, stored as https://api.example.com/#
```

**The last row is observation 7, and it predates this PR.** Go cannot
represent the distinction:

```text
https://a.example.com/    Fragment=""  String()="https://a.example.com/"
https://a.example.com/#   Fragment=""  String()="https://a.example.com/"
https://a.example.com/#x  Fragment="x" String()="https://a.example.com/#x"
```

`url.Parse` collapses "no fragment" and "empty fragment", so `if
u.Fragment != ""` in `validateResourceParameter`
(`coderd/oauth2provider/tokens.go:591`) cannot see a trailing `#`. RFC
3986 §3.5 permits a zero-length fragment, which RFC 8707 §2's "MUST NOT
include a fragment component" reads as forbidding. Validation parses the
value but persistence stores the raw string, so the `#` reaches
`resource_uri` and the stored audience is not textually equal to the
fragment-free form. A client appending a harmless-looking `#` would get
a token bound to an audience nothing matches.

The authorize-side call to `validateResourceParameter` is context in
this diff rather than an added line; this PR added the `invalid_target`
code for the failure.

</details>

<details>
<summary><b>24 and 25. The description cap and character
set</b></summary>

The obvious probe does not exercise the cap: a 4000-character
`code_challenge` produces a 116-character description, because that
message is fixed text and never quotes the offending value. The
descriptions that echo caller input come from a validator's own error.

| Request | `error` | Description length | Truncated | `Location` length
|
|---|---|---|---|---|
| `code_challenge_method` = 4000 chars | `invalid_request` | 2060 |
**yes** | 2176 |
| `scope` = 4000 chars | `invalid_scope` | 2060 | **yes** | 2174 |
| `code_challenge_method` = 2000 chars | `invalid_request` | 2035 | no |
2147 |
| `code_challenge` = 4000 chars | `invalid_request` | 116 | no | 234 |

2060 is exactly `maxErrorDescription` plus `" (truncated)"`. Row three
is the boundary control, so the cap fires on length rather than on the
presence of echoed input. Measured from the `Location` header rather
than a log line, since the stated reason for the cap is that the header
must survive intermediary proxies; the longest observed was 2176 bytes.

Both verbs agree, the cap living in the shared `redirectAuthorizeError`:

```text
GET, long method : desc_len=2060 truncated=True
GET, long scope  : desc_len=2060 truncated=True
```

The character set is enforced independently, later, in `errorURL`:

```text
input:       pl"ain\back<newline>tab<tab>end~unicode
description: unsupported code_challenge_method: pl'ainback tab end~ nicode
chars outside the RFC 6749 §4.1.2.1 set: []
```

`"` becomes `'`, `\` is dropped so it cannot escape the rewritten quote,
and anything below 0x20 or above 0x7E becomes a space. Checked
programmatically rather than by eye. The RFC sets no length limit, so
the 2048 cap is policy; the character set is conformance.

</details>

<details>
<summary><b>26. The registered callback query, retained except reserved
names</b></summary>

`plat479-reserved` registers a callback carrying all four reserved names
plus two ordinary ones, and DCR stores it unchanged:

```text
http://localhost:9876/callback?tenant=acme&code=FAKECODE&error=stale&error_description=oldmsg&state=oldstate&keep=yes
```

Success response, no `redirect_uri` sent:

```text
code  = coder_WXwBpvYR5u_...
keep  = yes
state = 516abeea9976119eb33ddc10f0ca04ed
tenant = acme
```

Failure response, same app:

```text
error = invalid_request
error_description = Invalid query params: code_challenge: must be 43 to 128 characters...
keep  = yes
state = 3a17bdd75dfcf740bda36124cd161fb7
tenant = acme
```

Stale registered values in either response: `FAKECODE` 0, `stale` 0,
`oldmsg` 0, `oldstate` 0.

Without the deletion step a registered `error=stale` would ride out on
the **success** response, and a client reading `error` before `code`,
the conventional order, would discard a valid authorization code.
`FAKECODE` is the mirror image on a failure response.

The cancel link obeys the same rule, being built by the same `withQuery`
path:

```text
http://localhost:9876/callback?error=access_denied&error_description=The+resource+owner...&keep=yes&state=e15b5f0f...&tenant=acme
```

RFC 6749 §3.1.2 requires the registered query be retained when adding
parameters, which is the `tenant` and `keep` half. It does not say what
to do when the registered query collides with the response parameters
§4.1.2 adds, so dropping the registered copies is the resolution that
keeps the client's read unambiguous.

</details>

<details>
<summary><b>27 and 28. Scope rejections, and declining
consent</b></summary>

`plat479-ci` carries the allowlist `coder:workspaces.access`.

```text
scope=template:update (outside)  -> error = invalid_scope
                                    'template:update': scope requests permissions beyond this app's allowed scopes
GET, same request                -> 302, 225 bytes, 0 Allow buttons, no consent page
scope=workspace:ssh (covered)    -> code issued, persisted scope workspace:ssh
GET, same request                -> 200, 5182 bytes, consent page naming the scope
```

`workspace:ssh` is granted although the allowlist names the composite
rather than that scope, which is the coverage-not-spelling rule from
#28045. This run only checks that the changes here left it delivering
the same answers.

The consent page for a narrow scope reads differently from the
unrestricted one: "to access your **admin** account with these
permissions?", the scope as `<li role="listitem">workspace:ssh`, and the
caution "These are technical permission names. Grant them only to an
application you trust."

The scope description shape differs from the parser's, coming from
`scopeFailureResponse` rather than the field join, which is the third
producer behind observation 1.

Declining consent, clicked in a real browser:

```text
error             = access_denied
error_description = The resource owner or authorization server denied the request
state             = a8bd538b365b61f54586892c73968b80
```

The page carries exactly one `href` and it is the cancel link. The
`state` matches, there is no `code`, and the code count for the app is
unchanged, the link being a plain `GET` to the client's callback that
never reaches `ProcessAuthorize`. The description is RFC 6749 §4.1.2.1's
own definition of `access_denied`, word for word.

That shared construction is what makes the `#nosec G203` annotation on
`CancelURI` sound: the URL is injected as a trusted `htmltemplate.URL`,
safe only because `newAuthorizeResponse` validated the registered scheme
before the response object could exist. Scenario 12 is the other half,
an app with a rejected scheme never reaching this page at all.

</details>

<details>
<summary><b>Where the consent page renders, across every GET
case</b></summary>

Only `GET` can render anything, and body size identifies which of four
renderers ran, so this is the table to check when a status code alone
looks right.

| Case | Status | Body bytes | Allow buttons | Rendered |
|---|---|---|---|---|
| well formed | 200 | 4928 | 1 | **consent page** |
| bad `code_challenge` | 302 | 263 | 0 | redirect stub |
| two bad fields | 302 | 319 | 0 | redirect stub |
| mismatched `redirect_uri` | 400 | 3814 | 0 | 400 error page |
| unparseable `redirect_uri` | 400 | 3846 | 0 | 400 error page |
| repeated `client_id` | 400 | 3846 | 0 | 400 error page |
| `client_id` in form body only | 400 | 77 | 0 | `httpmw` JSON, unstyled
|
| `{UPPERCASE}` `client_id` | 200 | 4940 | 1 | **consent page** |
| repeated `state` | 302 | 239 | 0 | redirect stub |
| corrupt callback, blocked scheme | 500 | 3346 | 0 | 500 error page |
| corrupt callback, unparseable | 500 | 3346 | 0 | 500 error page |
| `response_type=token` | 302 | 187 | 0 | redirect stub |
| `banana` / `code token` / `CODE` | 302 | 187 | 0 | redirect stub |
| empty `response_type` | 302 | 243 | 0 | redirect stub |
| `token`, no PKCE | 302 | 187 | 0 | redirect stub |
| `code`, no PKCE | 302 | 245 | 0 | redirect stub |
| unknown params, valid request | 200 | 4932 | 1 | **consent page** |
| `redirect_url` typo, valid request | 200 | 4932 | 1 | **consent page**
|
| `resource` malformed | 302 | 214 | 0 | redirect stub |
| `resource` valid absolute URI | 200 | 4932 | 1 | **consent page** |
| `resource=.../#` empty fragment | 200 | 4932 | 1 | **consent page**,
observation 7 |
| reserved params in registered callback | 200 | 4973 | 1 | **consent
page** |
| `scope` outside allowlist | 302 | 225 | 0 | redirect stub |
| `scope` covered by allowlist | 200 | 5182 | 1 | **consent page** |

The invariant: the page renders if and only if `extractAuthorizeParams`
returned no failure, so no input asks the owner to approve a request the
server has already refused. The `evil.example` rows are worth reading
carefully rather than alarming: RFC 6749 §3.1 requires a case variant
and a typo to be ignored, so from the server's view nothing was wrong
with either request, and both cancel links target the registered
callback.

About 4930 bytes is the consent page, about 3800 the 400 page, 3346 the
500 page, 77 the unstyled `httpmw` JSON, and low hundreds is Go's
redirect stub, whose length tracks the `Location` it embeds. Every
`unsupported_response_type` answer is exactly 187 bytes because they
share one description. Small variations within a renderer are expected:
the consent page moves 4928 to 4940 with a fresh CSRF token and the app
name.

As a regression signal, body size is more sensitive than the status
line. A change that rendered a page alongside a redirect would still
report 302, but the stub would come back at thousands of bytes rather
than hundreds.

</details>

<details>
<summary><b>Cleanup</b></summary>

```bash
for id in $(pg "SELECT id FROM oauth2_provider_apps WHERE name LIKE 'plat479%';"); do
  curl -s -X DELETE "$BASE_URL/api/v2/oauth2-provider/apps/$id" -H "$AUTH_HEADER" \
    -o /dev/null -w "$id -> %{http_code}\n"
done
pg "SELECT count(*) FROM oauth2_provider_apps WHERE name LIKE 'plat479%';"
./scripts/coder-dev.sh oauth2-provider dcr disable
```

All planted callback rows were restored before deletion, so no fixture
was left holding a value DCR would refuse.

</details>

---------

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: Steven Masley <Emyrk@users.noreply.github.com>
2026-09-03 10:54:34 -07:00
e035967ef1 feat: mint access tokens with the negotiated scope (#28237)
**TL;DR**

First of PLAT-480, continuing the merged PLAT-479 stack. #28178 decided
what the scope is and #28179 reported it; this PR is the one that makes
it bind. Until now the scope negotiated at `/oauth2/authorize` was
recorded on the grant but never reached the token: every OAuth2 access
key was minted with `coder:all`, so a client that asked for
`coder:workspaces.access` still got full API access.

| PR | What it does |
|---|---|
| #28167 | `ScopesCover` compares an allowlist against a request by
permission coverage rather than by name. Merged. |
| #28178 | The authorize endpoint checks the request against the scope
catalog, grants it only if the app's filtered allowlist covers it, and
persists the result on the code. Merged. |
| #28179 | The consent page lists the negotiated permissions, and
`invalid_scope` reaches the client's own callback per RFC 6749 §4.1.2.1.
Merged. |
| **#28237 (this)** | The exchange mints the API key with the stored
scope, and the token response states what was granted. |
| #28740 | Re-checks the app allowlist at redemption, for a code whose
app was narrowed after the code was issued. |
| #28751 | Lets a refresh narrow the granted scope, which is the `scope`
request parameter this PR still ignores. |

- #28179 has merged, so this diff is against main and reviews on its
own.
- **This is the first PR in the stack where a token's authority actually
changes.** Everything before it recorded and reported. That is the blast
radius worth reviewing: a grant that negotiated a narrow scope now gets
a key that is genuinely narrower, and anything that was quietly relying
on `coder:all` will start seeing 403.
- Grants that predate the scope columns carry what migration 000569
backfilled, `coder:all`, so they refresh unrestricted and reach what
they reached before. Pinned by `BackfilledScopeRefreshesUnrestricted`,
seeded the way the migration leaves a row rather than exchanged.
- The actor that writes the key stays `rbac.ScopeAll`. Narrowing it to
the granted scope would deny `api_key:create` and fail the exchange. The
bound lives on `api_keys.scopes`, which dbauthz reads on every request,
not on the actor doing the insert.

**Where in the flow**

<details>
<summary>Diagram: where the stored scope turns into a bound on the
token</summary>

```mermaid
flowchart TD
    AZ["GET or POST /oauth2/authorize<br/>negotiates the scope and persists it on the code<br/>PR 28178, merged"]
    AZ --> TK["POST /oauth2/tokens"]

    TK --> GT{"grant_type"}
    GT -->|authorization_code| S1["scope read from codes.scope"]
    GT -->|refresh_token| S2["scope read from tokens.scope<br/>req.Scope is parsed and ignored, PR 28751"]

    S1 --> CV["scopeStringToAPIKeyScopes, this PR<br/>every name must be in the api_key_scope enum<br/>an empty list is an error, not coder:all"]
    S2 --> CV

    CV -->|"unknown name, or empty"| ERR["400 invalid_grant, this PR<br/>error_description names the offending value"]
    CV -->|mintable| GEN["apikey.Generate with Scopes, this PR<br/>was: no Scopes, so every key defaulted to coder:all"]

    GEN --> KEY["api_keys.scopes<br/>read by dbauthz on every later request:<br/>this is where the grant becomes a bound"]
    GEN --> ROW["the new tokens row carries the same scope,<br/>so the next refresh reads back the same grant"]

    KEY --> RESP["200 with scope in the token response, this PR<br/>RFC 6749 5.1, OAuth 2.1 1.4.1"]
    ROW --> RESP
```

</details>

- Token endpoint only, both grants. Authorize is untouched apart from
two comments that said the scope was not yet enforced.
- `scopeStringToAPIKeyScopes` turns the stored scope string into
`database.APIKeyScopes` for `apikey.Generate`. Names are checked there
rather than inside `apikey.Generate`, whose error would surface as a 500
rather than an OAuth2 error.
- An empty list is an error, not an unrestricted key. `apikey.Generate`
defaults an empty scope list to `coder:all`, so an empty stored scope
would silently widen the grant instead of failing it. The same reasoning
is why the `expandRBACScope` comment in `modelmethods.go` was stale:
that function already errored on an empty list while its comment still
promised `rbac.ScopeAll`.
- A stored scope outside the `api_key_scope` enum answers 400
`invalid_grant` rather than 500, with the offending name in
`error_description`. Not `invalid_scope`: RFC 6749 §5.2 and OAuth 2.1
§3.2.4 both scope that code to what the client requested, and this value
is server state the client cannot change by asking differently. The
grant is what is unusable, and re-authorizing is the only way out.
- Refresh copies the presented token's scope onto both the new API key
and the rotated refresh row. `req.Scope` is parsed but ignored; #28751
is where it starts being honoured.
- The swagger annotation moves from `oauth2.Token` to
`codersdk.OAuth2TokenResponse`, because `oauth2.Token` has no `scope`
field to document.
- `EveryCatalogNameMintable` pins the catalog against the
`api_key_scope` enum. The two are maintained separately, and a name
negotiable at authorization but unmintable at exchange would leave a
client holding a code it can never redeem.

**What it satisfies**

Checked against both RFC 6749 and the OAuth 2.1 draft
(draft-ietf-oauth-v2-1-15). The 2.1 draft moves the scope-reporting rule
from RFC 6749 §5.1 into §1.4.1 and grades it in §3.2.3, but the
requirement is the same one.

- RFC 6749 §5.1, OAuth 2.1 §1.4.1 — "If the issued access token scope is
different from the one requested by the client, the authorization server
MUST include the `scope` response parameter in the token response".
OAuth 2.1 §3.2.3 grades `scope` as "RECOMMENDED, if identical to the
scope requested by the client; otherwise, REQUIRED". Both grants set it
unconditionally, which covers both grades. Without it, a client whose
request was narrowed against the app's allowlist, or defaulted because
it requested nothing, first learns its bounds from an unexplained 403.
Pinned by `IssuedTokenBoundsTheAPI`, which requests no scope and asserts
the response names what it got.
- OAuth 2.1 §1.4.1 — "The authorization server MAY fully or partially
ignore the scope requested by the client, based on the authorization
server policy or the resource owner's instructions." This is what
licenses ignoring `scope` on a refresh, and the licence is conditional
on reporting the actual grant. Before this PR the exchange both ignored
the request and said nothing, which is the half that was not defensible.
- OAuth 2.1 §4.3.1, RFC 6749 §6 — the refresh `scope` parameter is
OPTIONAL and "if omitted is treated as equal to the scope originally
granted by the resource owner". Exactly what `refreshTokenGrant` does
with `dbToken.Scope`.
- OAuth 2.1 §4.3.3 — "If a new refresh token is issued, the refresh
token scope MUST be identical to that of the refresh token included by
the client in the request." Rotation copies `dbToken.Scope` onto the new
refresh row.
- OAuth 2.1 §3.2.3 — "If refresh tokens are issued, those refresh tokens
MUST be bound to the scope and resource servers as consented by the
resource owner. This is to prevent privilege escalation by the
legitimate client". The refresh row has carried the negotiated scope
since #28178; this PR is what makes the access token minted from it
carry the same bound. Pinned by `RefreshDoesNotWidenTheScope`.
- OAuth 2.1 §1.4.1 — a client that omits `scope` must be met with "a
pre-defined default value or fail the request indicating an invalid
scope", and the server "SHOULD document its scope requirements and
default value". #28178 chose the default; `scopes_supported` and the
integration guide document it. Unchanged here, but it is why
`IssuedTokenBoundsTheAPI` can request nothing and still receive a
bounded token.
- Not yet: OAuth 2.1 §4.3.1 also says "The requested scope MUST NOT
include any scope not originally granted by the resource owner." A
refresh naming a narrower scope is ignored rather than honoured, and one
naming a wider scope is ignored rather than rejected with
`invalid_scope` (§3.2.4, "exceeds the scope granted by the resource
owner"). Neither case widens a token, so there is no escalation, and
§1.4.1's MAY plus the `scope` response parameter keep the client
correctly informed in the meantime. #28751 is where narrowing lands.
Documented under Limitations until then.

---

First of three, [PLAT-480](https://linear.app/codercom/issue/PLAT-480).
Depends on #28179, which has merged, so this reviews against main.

#28178 decided the scope and #28179 reported it. This PR is the one that
makes a token feel it.

---------

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: Steven Masley <Emyrk@users.noreply.github.com>
2026-09-02 13:14:44 -07:00
Jeremy RuppelandCoder Agents 8b8c80614d feat: expose template builder agents and per-module agent (#28905)
Part of [DEVEX-556](https://linear.app/codercom/issue/DEVEX-556). Builds
on the base agent metadata added in #28794.

## Summary

Surfaces the backend multi-agent support through the SDK and HTTP API:
the bases endpoint returns each base's agents, and
compose/create-template accept a per-module `agent_name`.

## What changed

- `codersdk/templatebuilder.go`: added `TemplateBuilderBaseAgent`,
`TemplateBuilderBase.Agents`, and
`TemplateBuilderComposeModule.AgentName` (`agent_name,omitempty`).
- `coderd/templatebuilder_handler.go`: added `baseAgentsToSDK`; the
bases endpoint returns agents, and the compose and create-template
endpoints forward `AgentName`.
- `make gen` regenerated `site/src/api/typesGenerated.ts`,
`coderd/apidoc/docs.go`, `coderd/apidoc/swagger.json`,
`docs/reference/api/schemas.md`, and
`docs/reference/api/templatebuilder.md`.
- `wizardState.test.ts`: added `agents: []` to the base fixture now that
the field is required on the generated type.

## Scope

SDK/API plumbing only. `ComposeModule.AgentName` now has a producer (the
handlers) but no caller sets it yet — the wizard UI (follow-up) is what
will populate it. No DB change, so `enterprise/audit/table.go` is
unaffected.

## Testing

- Bases endpoint asserts each base returns agents with exactly one
default (and the `dev`/`main` names).
- Per-module `agent_name` round-trip and unknown-agent (400) compose
tests.
- `make gen` (idempotent), `make fmt` (no changes), `golangci-lint
./coderd/... ./codersdk/...`, the Go template-builder tests, frontend
`tsc` + `biome check`, and docs markdownlint all pass.

<details>
<summary>Design note</summary>

`display_name` is intentionally returned as `""` for now: every base is
single-agent, so no base authors a per-agent label. The field exists for
multi-agent bases where the wizard radio needs a human-readable name;
consumers fall back to the agent `name` when it is empty.

</details>

---
🤖 Generated by Coder Agents on behalf of @jeremyruppel.

Co-authored-by: Coder Agents <agents@coder.com>
2026-09-02 12:51:33 -04:00
dylanhuff-at-coder 4875ae218a feat: enforce user secret file policy (#28518)
Adds the deployment option `--user-secrets-disable-file-path` /
`CODER_USER_SECRETS_DISABLE_FILE_PATH` / `userSecretsDisableFilePath`
(default `false`). When it is set, the create and batch-create
user-secret endpoints reject requests with a non-empty file path, and
PATCH validates the post-update state under a `SELECT ... FOR UPDATE`
row lock, rejecting a new or changed file path as well as enabled states
that would have no effective delivery target. Legacy stored paths are
preserved, so users can still edit, disable, clear the path, or migrate
the secret to env delivery.

This is PR 1 of a 4-PR stack; follow-ups cover manifest filtering, the
dashboard, and CLI/docs.

Refs #27488

Reviewed and updated by Coder Agents on behalf of @dylanhuff-at-coder.
2026-09-02 09:21:36 -07:00
Bartek Gatz 55340de360 feat: add deployment setting to disable the Codernauts game (#27662)
Adds a `codernauts_enabled` runtime setting (default: enabled) that
controls whether the Codernauts game link appears in the user dropdown
menu. A toggle at the bottom of the Deployment > Appearance page
controls it, and the value persists in `site_configs`, so it survives
restarts and upgrades without any CLI flags or environment variables.

The setting works on all licenses: `GET /api/v2/appearance` reports it
from both the enterprise and the default (AGPL) appearance fetchers, and
`PUT /api/v2/appearance` persists it gated only by the deployment-config
RBAC permission, not by an entitlement. On the Appearance page the
toggle renders outside the Premium paywall introduced in #27948; the
other appearance fields keep their existing licensing behavior.

Note: a pure AGPL-only build of coderd does not register the
`/api/v2/appearance` routes at all, so the toggle cannot be saved there.
The standard `coder server` binary is unaffected regardless of license
state.

<details>
<summary>Implementation notes</summary>

- `coderd/database/queries/siteconfig.sql`: new `GetCodernautsEnabled` /
`UpsertCodernautsEnabled` queries backed by a `codernauts_enabled` key
in `site_configs`; defaults to `true` when unset (no migration needed).
- `coderd/database/dbauthz`: read has no authz checks (matching other
appearance reads); write requires `ResourceDeploymentConfig` update.
Coverage added to `dbauthz_test.go`.
- `codersdk`: `codernauts_enabled` added to `AppearanceConfig` and
`UpdateAppearanceConfig`.
- `coderd/appearance`: the default fetcher takes a `database.Store` and
reads the setting from the database so unlicensed deployments serve it
too.
- `enterprise/coderd/appearance.go`: fetches the value in `Fetch` and
persists it in `putAppearance`.
- Frontend: "Codernauts game" toggle section on Deployment > Appearance
saving immediately via the existing appearance mutation;
`codernauts_enabled` threaded from `useDashboard().appearance` through
Navbar to `UserDropdownContent`, which renders the menu item only when
enabled. Storybook stories with `play` functions cover toggling
(entitled and not) and the dropdown hiding the link when disabled.

</details>

---

🤖 This PR was generated by Coder Agents on behalf of @bartekgatzcoder.
2026-09-02 11:29:35 +02:00
Michael Suchacz d6b67e8e31 feat: enforce SSRF protection for MCP config-directed traffic (#28242)
Routes all MCP config-directed traffic from coderd and chatd through a
shared SSRF-protected HTTP client, now that organization admins, not
only deployment admins, control MCP server URLs (#27942 and its stack
below).

## Summary

- Uses the [`coder/safedial`](https://github.com/coder/safedial)
library: it blocks private and special-purpose destinations at dial time
(validating resolved addresses at connection time so DNS rebinding
cannot bypass the check) and rejects cross-origin redirects.
- Covers the complete traffic surface: OAuth2 discovery, dynamic client
registration, code exchange, token refresh, revocation, and runtime MCP
connections from chatd.
- Deployments that intentionally host internal MCP servers opt in via
the new `--mcp-allowed-private-cidrs`
(`CODER_MCP_ALLOWED_PRIVATE_CIDRS`) option.
- Includes the deployment configuration surface, generated docs, CLI
goldens, TypeScript types, and regression coverage for each traffic
path.

## Merge window

This protection originally lived inside #27942 and was split out to keep
that diff reviewable. Until this PR lands, the stack below ships with
only main's existing discovery IP-range guard
(`CODER_MCP_OAUTH2_DISCOVERY_ALLOWED_IP_RANGES`), while org admins can
already point MCP configs at arbitrary URLs. This PR should merge
promptly after the stack below it.

Top of the MCP org-separation stack (CODAGT-711 -> CODAGT-717 audit ->
CODAGT-712 ACLs -> CODAGT-806 token RBAC -> CODAGT-714 org picker ->
this PR).

> Mux (AI agent) authored this PR on Mike's behalf.

<!-- mux-attribution: model=claude-fable-5 thinking=high -->
2026-08-30 13:09:17 +02:00
Hank Hwang c9dedc65e6 fix: model nullable UUID fields as uuid strings in Swagger (#28684)
## TLDR
Updated `.swaggo` file to include new rule to replace all
`uuid.NullUUID` with a `type: string` and `format: uuid` in
`swagger.json`. One such property example is `context_file_agent_id`

## What

Nullable UUID fields (`uuid.NullUUID`) were rendered in the
Swagger/OpenAPI spec as nested objects (with `UUID` and `Valid`
sub-fields) instead of as a plain UUID string. This adds a swaggo
`replace` directive so `uuid.NullUUID` is modeled as a `string` with
uuid format, matching how these fields serialize over the wire and how
other UUID fields are already documented.

## Changes

- `.swaggo`: add `replace github.com/google/uuid.NullUUID string` (with
an explanatory comment), alongside the existing `NullTime` replacement.
- `coderd/apidoc/docs.go` and `coderd/apidoc/swagger.json`: regenerated
output reflecting the new modeling.

## Testing

- Generated docs via the standard `make gen` flow; `docs.go` and
`swagger.json` are the regenerated artifacts.

---

> This PR was created by Coder Agents on behalf of @hwang251.
2026-08-27 09:13:33 -07:00
Rowan SmithandClaude Opus 5 4830010c8e chore: update tailscale fork version used by coder to remove hairpin probes (#28682)
Updates the version of tailscale used by Coder to pull in
coder/tailscale/pull/132 in order to remove hairpin probes, a feature
long ago removed from upstream tailscale. Requested by a customer.

https://linear.app/codercom/issue/PLAT-537

---------

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 14:40:37 +02:00
EthanandMichael Suchacz d0e22343f0 fix: make chat model sharing work for sharers without directory access (#28542)
## Problem

This is a bug fix. Chat model sharing UAT (main @ 3c32408) found two
bugs when a delegated "Model Sharer" role holds only
`chat_model_config:read` + `chat_model_config:share`:

This state is reachable entirely through the UI in a normal
least-privilege setup:

1. An org admin restricts **Organization Settings → Workspace Sharing**
to `none` or `service_accounts`.
2. Under **Organization → Roles**, they create a custom role with only
the advanced permissions `chat_model_config:read` and
`chat_model_config:share`.
3. Under **Organization → Members**, they assign that role to the person
responsible for managing model access.
4. An admin creates a chat model and may initially share it with a
group.
5. The delegated sharer signs in, opens **AI Settings → Models → [model]
→ Share model**, where the failures below occur.

- **UAT-05 / UAT-08:** with workspace sharing set to `none`, the sharing
dialog errored out entirely ("Resource not found or you do not have
access to this resource"). The sharer couldn't view, add, or even remove
grants.
- **UAT-04:** with workspace sharing set to `service_accounts`, existing
group grants rendered as raw UUIDs, and groups didn't appear in the
add-principal search.

Both have the same cause. The ACL GET returned bare UUID→role maps, so
to display names (and to power its add-principal search) the dialog
called the generic org directory APIs: list organization members and
list groups. Those APIs require `organization_member:read` and
`group:read`, which the sharer role doesn't have. Ordinary members only
get them as a side effect of a *different* feature:
`OrgMemberPermissions` grants member read when the org's workspace
sharing mode isn't `none` (so users can pick coworkers to share a
workspace with), and group read only when it's `everyone`. The chat
dialog was silently borrowing those workspace-sharing permissions, so it
only worked when that unrelated org setting happened to grant them:

- `none`: no member read → dbauthz rejects the member listing → the
dialog's query chain 404s and everything blanks (UAT-05/08).
- `service_accounts`: member read but no group read → the group listing
silently filters to empty → group grants can't be resolved past their
UUIDs (UAT-04).

In short: an org setting about who can share workspaces was deciding
whether a delegated sharer could manage a chat model ACL.

## Fix

Stop borrowing directory permissions entirely; authorize everything
through the model's own share permission, using the two patterns the
repo already has for ACL editors:

- **Hydrate the ACL GET** to return resolved users and groups instead of
UUID maps, like `workspaceACL` and the per-chat ACL already do. Fixes
the raw UUIDs.
- **Add `GET .../acl/available`** returning assignable org members and
groups for the autocomplete, modeled directly on
`templateAvailablePermissions` (same query semantics, same
share-gate-then-system-context lookup that every sibling ACL endpoint
uses). Fixes candidate discovery. The new endpoint exists only because
the dialog has no permission-safe way to enumerate principals today; it
is plumbing for the fix, not a new feature.

This is exactly how template sharing already solves the same problem:
`templateACL` returns hydrated principals, and
`templateAvailablePermissions` gates on the template then does the
lookups under a system context because, per its own comment, "the caller
might not have permission to read all users". We differ from the
template version only where newer conventions exist:

| | Template | This PR |
|---|---|---|
| Gate | `ActionUpdate` on the template (predates `share`) |
`ActionShare` on the model, matching the workspace/chat/MCP ACL
endpoints |
| User candidates | Site-wide `GetUsers` | Non-system members of the
model's org only |
| Group hydration | Full member rosters | Member counts, batched in one
query |

The dialog now runs off the hydrated ACL plus the new endpoint, so all
three sharing modes take the same code path, and a discovery failure no
longer blanks existing grants. `OrgMemberPermissions`, the directory
APIs, and the PATCH (still a sparse UUID→role delta) are untouched.

## Diff breakdown

+1274/−191, but only about a third is product code:

- **Product (~450):** the two endpoints and hydration helpers
(`exp_chats_model_acl.go`, ~240), the dialog rework plus a new colocated
autocomplete component (~200), and SDK/site API plumbing (~60).
- **Tests (~625):** backend tests updated to hydrated shapes plus new
tests that reproduce the two failing UAT sharing modes end-to-end
(+418), rewritten and new story play tests (+198), and query/mock
updates.
- **Generated (~200):** swagger, apidocs, `typesGenerated.ts`, API
reference docs.

Depends on #28498 (stacked on the `/api/v2` chat API promotion; the new
endpoint is registered under both prefixes like its siblings).

---------

Co-authored-by: Michael Suchacz <203725896+ibetitsmike@users.noreply.github.com>
2026-08-27 10:49:13 +01:00
Susana Ferreira 276d0a8586 fix: expose effective member AI budgets (#28410)
## Description

The group members page could not distinguish an unlimited Everyone
fallback from a configured budget on the Everyone group when viewing
another group, so it incorrectly displayed `Unlimited` and `Everyone
(not allocated)` despite the configured Everyone budget.

The member spend response now includes `effective_budget` as the
canonical budget that currently applies to the user. The existing
`group_budget` field remains for backward compatibility and is
deprecated in favor of `effective_budget`.

## Changes

- Add `effective_budget` to group member AI spend responses.
- Deprecate `group_budget` in favor of `effective_budget`.
- Display budgeted Everyone as the governing group instead of
`Unlimited` and `Everyone (not allocated)`.
- Add database, API, conversion, and Storybook coverage.

Closes https://linear.app/codercom/issue/AIGOV-588

> [!NOTE]
> Generated by Coder Agents on behalf of @ssncferreira.
2026-08-26 16:17:07 +01:00
Michael Suchacz 351bb14032 feat: mount chat API routes under /api/v2 (#28496)
## Stack context

This is the base of a 3-PR stack promoting the chat API from
`/api/experimental` to `/api/v2`: server compatibility mounts (this PR),
codersdk promotion (#28497), and frontend path updates (#28498).

## Summary

Double-mount the stable chat and MCP handlers under `/api/v2` while
retaining the existing experimental routes for the one-release
compatibility window decided in CODAGT-921. CODAGT-922 tracks removing
the compatibility mounts.

The shared route builders preserve existing authentication and
middleware behavior. Experiment-gated, debug, tombstone, and legacy
default-organization model routes remain experimental-only. Signed file
URLs, external OAuth callback URLs, and mixed-version replica relays
also remain on the experimental prefix during the transition.

Update Swagger and the generated API reference for the promoted routes,
including the workspace lookup and a runnable raw-body chat file upload
example. Retain internal endpoints outside the published reference,
share chat-file rate limits across both prefixes, enable CORS for the v2
MCP routes, and cover dual mounts plus exclusions with compatibility
tests. Remote dogfood UAT passed for the promoted chat, model, MCP, and
file flows.

> [!NOTE]
> Xum acted on Mike's behalf in this pull request.
<!-- xum-attribution: model=claude-fable-5 thinking=high -->
2026-08-26 16:17:06 +02:00
McKayla はな 29995c9ef8 feat: add a per-template workspace rename setting (#27558) 2026-08-25 11:05:42 -06:00
Garrett Delfosse eedc0ece47 fix: avoid 500 when generating chat title from rename dialog and remove unused endpoint (#28306) 2026-08-25 06:52:45 -04:00
Jon Ayers 26de6140fb feat: add flag to disable workspace agent context sync (#28522) 2026-08-24 19:58:30 -05:00
Michael Suchacz b6d7653231 fix: apply MCP server selection when editing a chat message (#28471)
## Problem

Editing a user message in an Agent chat silently dropped the MCP server
selection. The composer renders the MCP picker in edit mode and toggles
update local client state, so the picker displayed the new selection,
but the edit request omitted `mcp_server_ids` at every layer (frontend
request builder, `codersdk.EditChatMessageRequest`, the `PATCH
/chats/{chat}/messages/{message}` handler, and `chatd.EditMessage`). The
chat's persisted selection never changed and the regenerated turn ran
without the newly enabled MCP tools. Two dogfood users hit this within
hours; there is no error anywhere and the UI shows the opposite of the
server state.

## Changes

- `codersdk`: add `MCPServerIDs *[]uuid.UUID` to
`EditChatMessageRequest`, mirroring `CreateChatMessageRequest` (nil
preserves the current selection).
- `chatd`: extract the send path's MCP update block into one shared
`applyRequestedMCPServerIDs` helper (explore-subagent snapshot
immutability guard plus Force On enforcement, Cure53 CDM-02-010) and
call it from both `SendMessage` and `EditMessage`, so enforcement cannot
drift between the two paths.
- `coderd`: extract the send handler's request validation (dedupe,
enabled-in-organization check, persisted-ID exemption) into
`normalizeRequestedChatMCPServerIDs` and wire it into the edit handler,
which now threads the selection into `EditMessageOptions`.
- Frontend: the edit request now includes `mcp_server_ids:
[...effectiveMCPServerIds]`, exactly like the send path, making the
picker's displayed state real.
- `make gen` artifacts (swagger, API docs, `typesGenerated.ts`).

## Tests

Each layer is covered and was proven with independent red toggles
(removing one layer's wiring fails only that layer's tests):

- chatd (`TestEditMessage_MCPServerIDs`): edit applies a provided
selection, nil preserves it, an emptied list cannot remove a `force_on`
server, and explore subagent chats keep the spawn-time snapshot.
- API (`TestPatchChatMessage/MCPServerIDsApplied`,
`MCPServerIDsInvalidRejected`): persistence via the endpoint, omission
preserves, unknown IDs get the same 400 as the send path.
- Storybook (`EditAppliesMCPServerSelection`): toggling a server on
during an edit puts it in the edit request payload.

Note: the `AgentChatPage.stories.tsx` story "Queued For Capacity After
Polling" fails locally on current main as well (verified against the
main baseline with this branch's changes reverted); it is unrelated to
this diff.

Remote dogfood UAT ran on the exact head and passed, including proof
that the regenerated turn actually gains the newly enabled MCP server's
tools.

> 🤖 Xum acted on Mike's (@ibetitsmike) behalf. • Model:
`anthropic:claude-fable-5`
<!-- xum-attribution: model=anthropic:claude-fable-5 -->
2026-08-24 17:39:52 +02:00
Cian Johnston 042421694f fix: surface duplicate AI provider hostname as warning, not error (#28208)
Two live `ai_providers` rows can share the same `base_url` hostname.
This is an intended configuration (AIGOV-97): the direct routing path
(`/api/v2/ai-gateway/{name}/...`) routes by provider name and fully
supports it. Only the MITM proxy path (`aibridgeproxyd`) cannot
disambiguate, because its `nameByHost` map is 1:1.

`classifyProviderRow` marked the second provider as
`ProviderStatusError`, which surfaced as a misleading log warning and a
`status=error` Prometheus label. The provider is not actually broken; it
is fully routable via the direct path. The error was only visible in
server logs and Prometheus, not in the API or UI.

This change:

- Extracts `BaseURLHostname` into `coderd/aibridged` and eliminates
three duplicate implementations (proxy classifier, chatprovider, and the
API layer now uses it for status checks).
- Downgrades the proxy classifier from `ProviderStatusError` to a new
`ProviderStatusProxyExcluded` status. The provider is not broken; it is
just not reachable via the MITM proxy path.
- Adds `Status *AIProviderStatus` to `codersdk.AIProvider` with
`Warnings []string`, matching the `WorkspaceProxy` precedent.
- Populates warnings on create, update, GET, and list when a provider's
hostname collides with another enabled provider.
- Adds an integration test proving that a duplicate provider is routable
via the direct path but not via the proxy path.

Refs AIGOV-596

Generated by Coder Agents



https://github.com/user-attachments/assets/bf323837-5121-4d1b-ab19-593f1d32374e
2026-08-24 13:27:22 +01:00
Ethan c275327fb7 feat: audit operational agent settings (#28369)
Depends on #28394.

Audit changes to seven operational agent settings through a typed
`chat_operational_settings` resource. Each handler serializes baseline
capture and mutation with a per-setting advisory lock, keeps both
operations in one transaction, and records field-specific diffs.

Effective no-op requests now skip both the database upsert and the
success audit, including absent rows whose effective value is already
the default. The change also retains the allowlisted raw `site_configs`
query, updates generated audit and API surfaces, adds failure and
concurrency coverage, and reuses the existing AuditPage resource-filter
story pattern.

Relates to CODAGT-720.
2026-08-24 11:53:47 +01:00
Michael SuchaczandEthan Dickson 0ef331fcd0 feat!: scope chat model override settings to organizations (#28442)
Moves chat model override settings from deployment scope to organization
scope, following the org-scoping of chat model configs in #28440.

Depends on #28440 (based on
`ethan/chat-models-availability-consolidation`).

Remote dogfood UAT passed on this change, including multi-organization
scenarios on a licensed deployment with real models.

## Problem

Chat model configs are now owned by organizations, but the override
settings that reference them stayed deployment-wide: admin
subagent/title/compaction overrides in `site_configs`, personal
overrides in `user_configs`, and the advisor model embedded in the
advisor runtime JSON. A single deployment-level override can only point
at one organization's model config, so overrides effectively worked for
the default organization only, and title generation had to soften its
hard-failure contract to tolerate unusable overrides.

## Fix

Admin overrides (general, explore, title generation, compaction,
advisor) and personal overrides (root, general, explore) become
organization-scoped rows in two new typed tables,
`chat_organization_model_overrides` and `chat_user_model_overrides`.
Composite foreign keys on `(organization_id, model_config_id)` make
cross-organization model references unrepresentable. The migration
deletes the legacy serialized `site_configs`/`user_configs` override
keys instead of migrating them; admins and users re-select models in the
new organization-scoped settings.

New endpoints live under
`/api/experimental/organizations/{organization}/chats/model-overrides`
(admin defaults) and
`/api/experimental/organizations/{organization}/members/{user}/chats/model-overrides`
(personal overrides); the legacy deployment-level routes are removed.
chatd resolves overrides via the chat's organization, restoring the
hard-failure contract for configured-but-unusable title generation
overrides. The deployment-level `AllowUsers` switch and advisor runtime
limits stay deployment-scoped.

The Coder Agents settings UI gains a per-organization "Defaults &
overrides" page under AI Settings > Models, and the personal overrides
page gains an organization selector for users in multiple organizations.
Advisor telemetry now reports per-organization override rows.

**BREAKING CHANGE**: Existing deployment-level overrides are dropped,
not migrated. Every organization starts with no overrides after upgrade,
and model selections must be re-applied in the new settings. The legacy
deployment-level override endpoints are removed.

Closes https://linear.app/coder/issue/CODAGT-872

> [!NOTE]
> Xum (AI agent) authored this PR on behalf of @ibetitsmike.

<!-- xum-attribution: model=claude-opus-4-6 thinking=high -->

---------

Co-authored-by: Ethan Dickson <ethan@coder.com>
2026-08-24 11:20:26 +01:00
Ethan 3809314b31 refactor!: consolidate chat model availability (#28440)
## Summary

Consolidate chat model management and runtime availability into a single
organization model collection contract. The organization model response
now includes provider availability, unavailable reasons, and
unsupported-provider details, so callers no longer need a separate
availability request.

Depends on #28439.

## Problem

The organization-scoped API exposes two overlapping representations of
the same chat models:

```text
GET /api/experimental/organizations/{organization}/chats/models
GET /api/experimental/organizations/{organization}/chats/models/available
```

The collection route returns organization-owned model configurations and
redacted provider descriptors. The availability route returns a second
catalog-oriented response with provider availability and synthetic
catalog entries. Callers must fetch, reconcile, and cache both responses
even though they describe the same effective set of models.

This creates duplicated server logic and separate SDK, OpenAPI, and
frontend types. It can also produce inconsistent client state when one
request succeeds while the other fails or when the two responses are
refreshed at different times.

## Fix

Return all model-management and runtime-availability information from
the organization model collection route:

```text
GET /api/experimental/organizations/{organization}/chats/models
```

`OrganizationChatModelsResponse` continues to return the caller-readable
organization model configurations and redacted provider descriptors.
Provider descriptors now also include:

- `available`
- `unavailable_reason`

The response additionally includes `unsupported_providers`, allowing
clients to explain configurations that the Agents harness cannot use.

Frontend model settings, agent creation, existing chat recovery, and
model override surfaces now consume this single collection query. This
keeps model ownership, caller visibility, provider availability, and
unsupported-provider guidance in one cache entry and one authorization
path.

The default-organization compatibility route remains available:

```text
GET /api/experimental/chats/models
```

It now returns the same consolidated `OrganizationChatModelsResponse`
contract instead of the former availability response.

## Breaking change

This removes the separate organization model availability endpoint:

```text
GET /api/experimental/organizations/{organization}/chats/models/available
```

Clients using that endpoint must migrate to:

```text
GET /api/experimental/organizations/{organization}/chats/models
```

This PR also removes the corresponding public experimental client and
schema surface:

- `ExperimentalClient.ChatModelAvailability`
- `ChatModelAvailabilityResponse`
- `ChatModelProvider`
- `ChatModelCatalogEntry`
- The generated OpenAPI schema and frontend API/query types for the
availability endpoint

Callers must use `ExperimentalClient.ChatModels` and
`OrganizationChatModelsResponse` instead. Availability is now reported
on each `ChatModelProviderDescriptor`, and unsupported providers are
returned in `OrganizationChatModelsResponse.UnsupportedProviders`.

The default-organization compatibility route retains its URL, but its
response schema changes from `ChatModelAvailabilityResponse` to
`OrganizationChatModelsResponse`. Direct HTTP clients and generated
clients that decode the previous response type must be updated even if
they do not use the removed organization-scoped `/available` route.
2026-08-24 11:20:26 +01:00
Ethan aefc761fcf feat!: enforce organization scope for chat models (#27959)
## Summary

Move chat model management and runtime availability to
organization-scoped API routes. The API now identifies the organization
in every model-management operation, applies chat model RBAC and ACLs
within that organization, and returns only models the caller can use
there.

Depends on #27958.

## Problem

Chat model configurations now belong to organizations, but the existing
experimental management and discovery APIs are deployment-shaped. Their
routes do not identify an organization, SDK methods do not accept an
organization ID, and item operations address a model configuration by ID
alone.

That contract cannot safely expose organization-owned model
configurations. It also does not provide the organization-specific
provider information, authorization behavior, ACL management, or
recovery guidance needed by the settings and chat clients later in the
stack.

## Fix

Add organization-scoped chat model endpoints under:

```text
/api/experimental/organizations/{organization}/chats/models
```

The new API provides:

- Organization-scoped model listing and creation.
- Model get, update, and delete operations scoped by both organization
and model ID.
- Model ACL read and update operations.
- Organization-scoped runtime availability and provider status.
- Redacted provider descriptors for model-management clients.
- Chat model RBAC and row ACL enforcement for writes, runtime discovery,
and Chatd model selection.
- Recovery guidance when a chat's organization has no usable local
model.

Historical chats can retain a model reference owned by another
organization, but runtime selection uses an authorized local default
when one is available. Foreign, inaccessible, disabled, and otherwise
unusable models are not exposed as usable selections for the
organization.

The existing collection and availability routes remain as
default-organization compatibility endpoints in this layer:

```text
GET  /api/experimental/chats/model-configs
POST /api/experimental/chats/model-configs
GET  /api/experimental/chats/models
```

These compatibility routes do not provide access to models in
non-default organizations. The previous unscoped item update and delete
routes are removed because item operations must include the owning
organization.

## Breaking change

This changes the experimental HTTP API and Go SDK contracts for chat
model management.

Clients must migrate model-management and availability requests to the
organization-scoped routes and provide an organization identifier. In
the Go SDK, the following methods now require an organization ID:

- `ChatModelAvailability`
- `ChatModels`
- `CreateChatModel`
- `UpdateChatModel`
- `DeleteChatModel`

`ChatModels` also changes its return value from `[]ChatModel` to
`OrganizationChatModelsResponse`, which includes the organization models
and the redacted provider descriptors required by management clients.
New item and ACL methods likewise require both the organization ID and
model ID.

The default-organization compatibility routes reduce the immediate
impact for collection and availability callers, but they do not preserve
the previous item API or SDK method signatures. Generated clients and
direct consumers of the experimental API must be updated before adopting
this change.

_This pull request description was generated by Coder Agents._
2026-08-24 11:20:25 +01:00
Ethan 9bd8899359 refactor: clarify chat model resource names (#27958)
Rename the admin-managed SDK resource from `ChatModelConfig` to
`ChatModel`. Rename catalog entries from `ChatModel` to
`ChatModelCatalogEntry` and replace `ChatModelsResponse` with
`ChatModelAvailabilityResponse`.

API routes and JSON fields do not change. Experimental SDK and OpenAPI
schema names do change.

Depends on #27957

_This pull request description was generated by Coder Agents._
2026-08-24 09:43:45 +00:00
Ethan cbe24a3da5 feat: add chat model config RBAC resource (#27957)
Add the `chat_model_config` RBAC resource, actions, and API key scopes.
This layer includes the scope migration and generated API types. Model
list queries apply token scopes, SQL authorization filters, and row
ACLs.

Write and default-management handlers keep deployment configuration
authorization until the next API layer replaces it.

Depends on #27956

_This pull request description was generated by Coder Agents._
2026-08-24 09:43:45 +00:00
david-fraley cb52f5339d feat: show a premium paywall on the external auth settings page (#28435) 2026-08-21 11:45:38 -05:00
david-fraley a2afb458c5 feat: add premium paywall conversion telemetry (#28425)
**TLDR:** When someone using free Coder taps a "Start trial for free"
button, we now write down which page they tapped it on, and whether they
went on to actually start the trial. That tells us which locked features
make people want Premium.

## Brief summary

- Records two things: the button being clicked, and a trial actually
being started.
- Each click gets a random ID that travels with the person to the trial
form, so we can tell who finished from who walked away.
- Records a short label for the page (like `appearance` or `audit_log`),
never the web address, so no customer names are collected.

#28226 added the trial form and has since merged, so this now targets
`main`.

## Description

Today the "Start trial for free" button just moves you to the Premium
page and nothing is recorded, so there is no way to know which feature
convinced someone.

This adds two events:

| Event | Happens when |
|---|---|
| `cta_click` | Someone clicks "Start trial for free" |
| `trial_signup` | A trial licence is actually issued |

The click generates a random ID. It rides along to the trial form and
gets saved on the signup, so a click and a signup can be matched up
later. Clicks with no matching signup are the people who gave up. Trials
started without a paywall (for example straight from the sidebar) are
labelled `direct`, so "no label" never gets confused with "broken".

A few small choices worth knowing:

- The ID is kept in browser session storage, not in the web address, so
the URL stays clean. It expires after 30 minutes.
- The trial signup event is recorded by the server, not the browser, so
nobody can fake a signup.
- The browser cannot say which kind of event it is sending. It can only
report a click; the server labels it.
- Only people allowed to start a trial are counted. The server rejects
the rest, which matters because a few paywalls show their button to
everyone.
- Nothing is recorded at all when telemetry is turned off.

There is deliberately no "paywall was shown" event. Every admin page
visit would report one, which is a lot of noise for little insight.

Every paywall in the [Premium Copy
Updates](https://www.notion.so/coderhq/Premium-Copy-Updates-3c1d579be59280a59557cabd54fd8a1f)
table has a label. The rows that do not are badges and alerts with no
button to click.

## Proof

Ran a local Coder with no licence and clicked the paywall button on the
Appearance page.

What the browser sent, and the reply:

```json
POST /api/v2/deployment/premium-funnel-events  ->  204 No Content
{"id":"92d96afa-f975-4789-b8a7-b1fd19d4df85","source":"appearance","variant":"premium"}
```

What the browser saved for the trial form, same ID:

```json
{"id":"92d96afa-f975-4789-b8a7-b1fd19d4df85","source":"appearance","createdAt":1787268520857}
```

The click then landed on the Premium page with the trial form.
Screenshots of the network panel, the saved value, and the Premium page
are attached below. The form itself was not submitted, since that would
contact the real licence server.

## Testing

- Backend tests: a click is recorded with the server's own label, an
unknown page label is refused, and a normal member is refused.
- Storybook tests: button clicked and ID saved on all three paywall
styles, plus the case where the viewer only sees guidance text and has
nothing to click.
- Unit tests for the saved ID: expiry, bad data, unknown page label, and
the `direct` fallback.
- `make gen` is clean, and lint, types, and formatting pass.

Two stories fail on `main` today and fail the same way here:
`ExternalAuthSettingsPageView > Page` and `PremiumPageView > No
License`. Neither file is touched by this branch; both were verified
failing on `main` with this branch checked out elsewhere.

## Follow-up

coder/coder-telemetry-server#45 saves these events for reporting. It
merges after this one.

<details>
<summary>Plan and decisions</summary>

**Goal:** find out which Premium paywall drives trial signups, and tell
abandons apart from completions.

**Options considered and dropped:**

1. Put `?from=appearance` in the URL. Rejected: messy for users.
2. Use router state. Rejected: lost if the page is refreshed.
3. Let the browser report the signup. Rejected: it could be faked or
lost during the redirect.
4. Work out the page from the URL. Rejected: URLs contain organization
and template names.
5. Report every paywall impression. Rejected: one event per admin page
visit is noise, and the click is the signal.
6. Separate event types and tables. Rejected: makes the matching query
harder for no gain.

**Event shape:**

```go
type PremiumFunnelEvent struct {
	ID            uuid.UUID // event ID; for a click this is also the matching ID
	EventType     string    // cta_click | trial_signup, set by the server
	Source        string    // appearance, audit_log, ..., direct
	Variant       string    // which paywall design was shown
	AttributionID uuid.UUID // the click this signup came from
	UserID        uuid.UUID
	CreatedAt     time.Time
}
```

**Where the code lives:** nothing under `components/` imports from
`modules/`, so the paywall components stay presentational and only gain
an `onCTAClick` prop. New wrappers in `modules/paywall/` own the
reporting, one per paywall style, each taking a fixed `source`.

**Self-review:** the `frontend-review` skill caught that reading
permissions inside the wrapper forced an auth context into
presentational page views, which broke their stories. Every call site
already passes `canViewPremium={permissions.viewAllLicenses}` and the
button only renders when that is true, so the wrapper no longer reads
permissions and the server stays the gate. It also flagged one
unattributed paywall on the Security page, now wired up as
`browser_only`.

</details>

> Generated with [Coder Agents](https://coder.com/agents) on behalf of
@david-fraley
2026-08-21 10:31:01 -04:00
Samuel VolinandMatt Vollmer f44a3b59d7 feat: Premium Page CTA updates and form handling (#28226)
* Premium page has form for signing up for a trial
* Input validation and normalization for trial form fields
* api endpoint definition to make this processes seamless 
* license page updates to access `?success=true` queryparam and confetti

<img width="1512" height="833" alt="Screenshot 2026-08-17 at 10 08
41 PM"
src="https://github.com/user-attachments/assets/1766e724-c63a-4da9-b2bd-a98e6b66cc9b"
/>

<img width="1510" height="874" alt="Aug-17-2026 22-12-38"
src="https://github.com/user-attachments/assets/3a2a0f35-1176-4290-aa0d-b3f7d66c15d6"
/>


```mermaid
sequenceDiagram
    autonumber
    actor Owner
    participant UI as UI<br/>TrialRequestForm + PremiumPage
    participant API as API<br/>api.ts + coderd middleware
    participant Coderd as Coderd<br/>postTrialLicense + trialer
    participant Licensor as Licensor<br/>v2-licensor + Postgres + pubsub + cache

    Note over UI: mutation.status = "idle"

    Owner->>UI: fills 10 fields, checks acknowledgement
    UI->>UI: Yup validate, strip "acknowledged", submit
    UI->>API: POST /api/v2/licenses/trial<br/>cookie + X-CSRF-TOKEN
    Note over UI: mutation.status = "pending"

    API->>API: verify CSRF, apiRateLimiter,<br/>apiKeyMiddleware sets actor
    API->>Coderd: dispatch

    Coderd->>Coderd: audit.InitRequest, authorize license:create,<br/>httpapi.Read validates, HasLicense is false
    Coderd->>Licensor: POST trial request<br/>deployment_id unspoofable, 10s deadline
    Licensor-->>Coderd: 200, raw signed JWT

    Coderd->>Coderd: ParseClaimsIgnoreNbf, uuid.Parse
    Coderd->>Licensor: InsertLicense, updateEntitlements,<br/>publish PubsubEventLicense, commit audit log
    Licensor-->>Coderd: database.License
    Coderd-->>API: 201 codersdk.License
    API-->>UI: 201 codersdk.License

    Note over UI: mutation.status = "success"
    UI->>Licensor: invalidate entitlements + licenses keys
    UI->>Owner: navigate /deployment/premium?success=true
    Note over Owner: confetti fires on the success param
```

---------

Co-authored-by: Matt Vollmer <matthewjvollmer@outlook.com>
2026-08-21 09:54:54 -04:00
Susana Ferreira 50ecd6959f feat: add a source filter to the AI model price list (#28329)
## Description

With prices from the price book and prices set through the API both in the table, `list` showed only the price in effect and gave no way to tell where it came from. There was also no way to see the price book's entry for a model that had been overridden, since the listing resolves to the custom price.

This adds a `source` column to the output and a `--source` filter for narrowing to one source, or to every row with `all`.

Note: `--source` selects rows rather than models. A model carrying both a price book price and a custom one appears under either value, so the two filtered listings do not sum to the unfiltered one. `--source all` reports both rows at once.

## Changes

- Add `Source` to `codersdk.AIModelPrice` and `AIModelPricesFilter`, and map it in `db2sdk`, which previously dropped it.
- Filter `GetAIModelPrices` by source before resolution, so the price book's row for an overridden model stays reachable.
- Accept `all` as a source, which reports every price a model holds instead of resolving to one.
- Accept a `source` query parameter on the list endpoint, rejected with a 400 when outside the enum.
- Add `--source` to the list command and a `source` column to its output.
- Document the column and the filter, and lead the price section with `coder exp ai-model-prices list`.
- Distinguish a deployment's default prices from the price book that ships with each release.

Closes https://linear.app/codercom/issue/AIGOV-593

> [!NOTE]
> Initially generated by Claude Opus 5, modified and reviewed by @ssncferreira
2026-08-20 11:56:58 +01:00
david-fraley 8203b2ebaf feat!: hide Coder Tasks behind the enable-ai-tasks flag (#28008) 2026-08-19 14:46:10 -05:00
Michael Suchacz 7ca7c30f40 feat: add group and user ACLs to MCP server configs (#27944)
Adds group and user ACLs to org-scoped MCP server configs so
organizations can restrict specific MCP servers to subsets of members,
mirroring the template ACL pattern.

## Summary

- Migration adds `group_acl`/`user_acl` JSONB columns (nested `{"<id>":
{"permissions": [...]}}` shape) and seeds every existing config with its
organization's Everyone group read entry (the Everyone group ID equals
the org ID), so member access is unchanged by default. Creation seeds
the same entry.
- The blanket org-member read grant from the base PR is replaced by ACL
evaluation: Rego requires org membership for every ACL grant, and
`GetAuthorizedMCPServerConfigs` compiles ACL-aware SQL filters.
- New `ActionShare` (org admins) gates `GET|PATCH
/api/experimental/organizations/{organization}/mcp-servers/{mcpserverconfig}/acl`
(nested under the organization like the rest of the config surface).
PATCH validates principals against the config's organization, merges
sparse updates under a row lock, stamps `updated_by`/`updated_at`, and
is audited as a Write with `Old` captured before authorization. A config
deleted concurrently between the middleware fetch and the locked
re-fetch is concealed as 404, matching the update and delete handlers.
ACL columns are tracked in the audit table.
- ACL management is available in all editions (no enterprise
entitlement), documented in the MCP servers page. Revoking an ACL does
not retro-strip already-selected configs from existing chats; new
selection is blocked at chat create.
- Force On respects the ACL: the forced set is loaded as the chat owner,
so a `force_on` server whose ACL denies the owner never attaches at
create, send, or generation time.
- No rolling-upgrade machinery: upgrades run in scheduled maintenance
downtime, so the migration only backfills existing rows; the API sets
the Everyone read grant explicitly on every insert.

Stacked on #27943. Part of the MCP org-separation stack.

Closes https://linear.app/codercom/issue/CODAGT-712

UAT: verified on a dogfood instance with two members and a custom group:
Everyone-seed default visibility, group grant with Everyone removal
(non-member loses list/fetch/selection), foreign-principal rejection,
non-admin share denial (audited 403 / concealed 404), user_acl restore,
and audited ACL diffs.

> Mux (AI agent) authored this PR on Mike's behalf.

<!-- mux-attribution: model=claude-fable-5 thinking=high -->
2026-08-19 19:25:18 +00:00
Michael Suchacz 299e72ad30 feat: audit MCP server config changes (#27943)
Adds enterprise audit logging for MCP server config create, update, and
delete, with strict secret redaction. MCP configs hold credentials
(OAuth2 client secrets, API keys, custom headers), so admin changes to
them need an audit trail.

## Summary

- `enterprise/audit/table.go` gains an `MCPServerConfig` entry
enumerating every column: `oauth2_client_secret`, `api_key_value`, and
`custom_headers` are `ActionSecret` (never appear in diffs); dbcrypt
`*_key_id` bookkeeping, IDs, and timestamps are ignored; the remaining
config fields, including the endpoint URL fields, are tracked so
auditors can see which endpoints a config points at.
- Type registration in `coderd/audit` (diff, request, resource target
with org attribution), `codersdk/audit.go`, and a `resource_type` enum
migration.
- Handlers wire `audit.InitRequest`: create records `New`; update and
delete record `Old` from the param middleware before the
write-authorization check, so a readable-but-not-writable caller
produces an audited 403 while read-denied callers stay concealed as
unaudited 404s.
- Tests: create/update/delete audit entries, write-denied and
delete-denied 403 auditing, cross-org concealment producing zero
entries, and a serializer-level regression test proving none of the
three secret classes can reach a serialized diff.
- Review round: MCP config audit entries link to
`/ai/settings/mcp-servers/{id}`, audit table comments are trimmed per
review, and a fault-injection test pins that a config row surviving a
failed post-discovery credential update still gets its creation audit
entry.

Stacked on #27942 (org-scoped MCP configs). Part of the MCP
org-separation stack.

Closes https://linear.app/codercom/issue/CODAGT-717

UAT: verified on a trial-licensed dogfood instance: audit entries for
the full CRUD lifecycle with correct actor/org/target, redacted secrets
in the update and OAuth2 create diffs, and a full plaintext scan of the
audit dump finding zero secret leaks.

> Mux (AI agent) authored this PR on Mike's behalf.

<!-- mux-attribution: model=claude-fable-5 thinking=high -->
2026-08-19 19:11:52 +00:00
Michael Suchacz f2bc9ab1f5 docs: complete swagger annotations for organization-scoped MCP routes (#28064)
Adds the missing swagger annotations for the eight organization-scoped
MCP server config routes introduced in #27942 and checks in the
regenerated API artifacts (`coderd/apidoc`, `docs/reference/api`). No
behavior changes: 58 hand-written annotation lines, the rest is
generated output.

## Stack context

Part of the MCP org-separation stack (CODAGT-711 org scope -> apidocs ->
hardening -> CODAGT-717 audit -> CODAGT-712 ACLs -> CODAGT-806 token
RBAC). Split out of #27942 to keep the core cutover reviewable; these
routes live under `/api/experimental`, where main already ships several
MCP handlers without annotations, so the base PR is consistent with
existing precedent until this lands.

Closes nothing on its own; documentation completion for CODAGT-711.

> Mux (AI agent) authored this PR on Mike's behalf.

<!-- mux-attribution: model=claude-fable-5 thinking=high -->
2026-08-19 18:36:25 +00:00
Michael SuchaczandMathias Fredriksson 443e3b9b80 feat!: org-scope MCP server configs with RBAC (#27942)
Moves `mcp_server_configs` from deployment scope to organization scope
so each organization fully controls the MCP servers its members can use
with Coder Agents.

## Summary

- Migration: adds `organization_id` (NOT NULL, FK) and keeps existing
rows as the default organization's originals with credentials intact.
Other organizations start with no MCP servers and configure their own;
nothing is copied across organizations. Chats outside the default
organization keep any now-cross-organization `mcp_server_ids` entries;
the runtime already ignores IDs that do not resolve in the chat's
organization, so no data rewrite is needed. Slug uniqueness becomes
`(organization_id, slug)`.
- RBAC: new org-scoped `ResourceMCPServerConfig` with regosql converter
and `GetAuthorizedMCPServerConfigs`; org admins get in-org CRUD, org
members get read (replaced by ACL evaluation in the follow-up ACL PR in
this stack).
- API: all config routes nest under the organization, matching
templates: `POST|GET
/api/experimental/organizations/{organization}/mcp-servers` and
`GET|PATCH|DELETE .../mcp-servers/{mcpserverconfig}` (plus
`oauth2/connect`), resolved by a read-only param middleware that
conceals read-denied and cross-organization access as 404. Two routes
stay on the frozen `/api/experimental/mcp/servers/{mcpServer}` block:
the OAuth2 callback (the redirect URI baked into existing AS-side client
registrations) and `oauth2/disconnect`, which must remain reachable by
users removed from the organization so they can still revoke their token
grant.
- Chat runtime: selection validation and generation resolve configs
strictly by IDs, enabled state, and the chat's organization in SQL;
requested duplicates are normalized; invalid or cross-org IDs are
rejected with the precise ID list. IDs already persisted on a chat are
exempt from message-time rejection so disabling a selected server never
blocks sends; generation skips servers that are no longer usable.
- Frontend: API layer and admin settings pages target the new endpoints.
The admin page manages the default organization's servers; the org
picker is tracked separately (CODAGT-714).

- Security hardening from review: OAuth user grants are additionally
bound to `oauth2_revocation_url` (changing it invalidates grants, and a
racing OAuth callback gets 409 instead of recreating a grant).
Stack-wide SSRF protection for MCP config-directed traffic was split
into its own PR at the top of this stack (#28242) to keep this diff
reviewable; this PR keeps main's existing discovery IP-range guard.
OAuth2 auto-discovery now completes before the config row is inserted: a
failed discovery persists nothing, and there is no provisional row that
concurrent updates could race against.

Two follow-up PRs in this stack were split out to keep this diff
reviewable: #28064 completes the swagger annotations for the moved
routes (main already ships these experimental MCP handlers unannotated),
and #28065 carries hardening fixes and regression pins on top of the
cutover.

- Force On enforcement (landed on main mid-review) is org-scoped: the
forced set is read per chat organization
(`GetForcedMCPServerConfigsByOrganization`), so another organization's
`force_on` server never attaches to a chat.

## Breaking changes (experimental API)

The MCP server config endpoints move from the deployment-scoped
`/api/experimental/mcp/servers` block to organization-nested paths:
`POST|GET /api/experimental/organizations/{organization}/mcp-servers`
and `GET|PATCH|DELETE
/api/experimental/organizations/{organization}/mcp-servers/{mcpserverconfig}`
(plus `oauth2/connect`). The old paths are removed, so API consumers
must supply an organization. Two routes intentionally stay on the frozen
`/api/experimental/mcp/servers/{mcpServer}` block: the OAuth2 callback
(its redirect URI is baked into existing AS-side client registrations)
and `oauth2/disconnect` (must remain reachable by users removed from the
organization). These endpoints are under `/api/experimental`, so no
deprecation window is provided.

## Rolling upgrades

During a rolling deploy, an old replica creating an MCP config can fail
the new `NOT NULL organization_id` constraint until it is upgraded
(reads are unaffected: old binaries' generated queries select their own
column lists). This matches the repo's existing precedent for additive
NOT NULL migrations (000562) and affects only the admin config-create
path in the upgrade window.

Upgrades are expected to run in scheduled maintenance downtime with the
database locked during migration, so the migration ships no
rolling-upgrade compatibility machinery. The down migration deletes
organization-created configs (their chat references are cleaned by the
000510 delete trigger) and restores deployment-wide slug uniqueness.

Part of the MCP org-separation stack (CODAGT-711 -> CODAGT-717 audit ->
CODAGT-712 ACLs -> CODAGT-806 token RBAC).

Closes https://linear.app/codercom/issue/CODAGT-711

UAT: validated end to end on a two-org dogfood deployment, including a
real pre-migration to post-migration upgrade, cross-org isolation
(404s), same-slug-two-orgs, chat selection gating, and a live MCP tool
call through the org-scoped generation path. The migration was later
revised to keep existing rows in the default organization only (no
per-organization copies); that revision is covered by the migration test
suite.

> Mux (AI agent) authored this PR on Mike's behalf.

<!-- mux-attribution: model=claude-fable-5 thinking=high -->

---------

Co-authored-by: Mathias Fredriksson <mafredri@gmail.com>
2026-08-19 18:04:13 +00:00
Jon Ayers 821d91fabd fix: log tailnet tunnel authorization decisions (#27819) 2026-08-18 18:21:50 -05:00
Mathias FredrikssonandMichael Suchacz d3f08b1983 feat: audit chat system instructions changes (#27668)
Adds an audit record for administrative events on the deployment-wide
chat instruction settings (system prompt, the include-default toggle,
and the plan-mode instructions), per CODAGT-719 and operator decision
D5. Each endpoint records under a stable identity: resource type
`chat_instruction_settings`, a fixed resource ID and a human-readable
target ("System prompt", "Plan mode instructions"), so two changes to
one setting share an ID and history-by-setting works. A real change
exports a Write entry with the old-to-new text visible; a
value-identical PUT still upserts and still returns 204 but records
nothing.

Attempts are recorded, not only transitions. Identity is assigned before
the authorization check, so a denied PUT exports a 403 row with an empty
diff (no request content reaches it), a validation failure exports a 400
row, and a write failure exports a 500 row, each with an empty diff; an
operator can tell "nothing changed" from "something changed and capture
degraded" by the status code.

The write path stays authoritative. The advisory lock and, on plan-mode,
the transaction exist only to serve change-detection; if any of that
machinery fails (lock, begin, commit, rollback), the handler runs main's
idempotent write path directly and derives the response from it, so a
member-visible failure of audit-only infrastructure can never replace
main's successful response. Accepted consequence: when the lock cannot
be taken, two concurrent identical writes can produce two rows instead
of one. That is audit degradation, which is allowed; changing a member's
response is not. Write failures keep the exact response the endpoint
produced before this wiring (transaction error for the system prompt,
which was always transactional; the raw write error for plan mode, which
was not), and the full transaction error is logged so rollback failures
cannot vanish.

<details>
<summary>CODAGT-66 plan entry: S1 (verbatim)</summary>

**S1 `feat: audit chat system instructions changes`** (CODAGT-719; base:
main)

- Struct: `database.ChatSystemPromptSettings{ID uuid.UUID; SystemPrompt
string; IncludeDefaultSystemPrompt bool; PlanModeInstructions string}`
in `coderd/database/types.go` (ticket-sketched shape; one struct, both
endpoints).
- Registration: union entry (diff.go), table.go entry (`id`
ActionIgnore, other three ActionTrack), `AuditActionMap` Write-only;
four request.go cases (`ResourceTarget` "", `ResourceID` from struct,
`ResourceType` new enum value `chat_system_prompt_settings`,
`ResourceRequiresOrgID` false with the "Artificial ID / deployment
singleton" comment convention).
- Migration: `ALTER TYPE resource_type ADD VALUE IF NOT EXISTS
'chat_system_prompt_settings';` comment-only no-op down (000558 shape);
number picked at push per the numbering constraint.
- codersdk: constant + prose `FriendlyString` ("chat system prompt
settings"); `TestAuditDBEnumsCovered` forces both. `coderd/audit.go`
presentation switches: rely on safe defaults (no link, generic
description); no FE changes (filter label falls back to capitalized
value; acceptable per precedent).
- Wiring `putChatSystemPrompt` and `putChatPlanModeInstructions`:
InitRequest with Action Write; artificial `ID: uuid.New()` on `New` only
when a change is detected; no-op suppression by leaving both aReq sides
unset (nil resource IDs skip the log, request.go skip rule); the write
path itself stays byte-identical (upserts still run unconditionally).
- `putChatSystemPrompt` (writes two keys conditionally in one existing
tx): inside that tx, read the pair via `GetChatSystemPromptConfig` for
`Old`, perform the conditional writes exactly as today, then RE-READ the
pair for `New`. The re-read is load-bearing:
`include_default_system_prompt` is computed from the toggle row AND the
prompt, so a prompt-only write can flip the effective value without the
request carrying the pointer. `PlanModeInstructions` stays zero on both
sides.
- `putChatPlanModeInstructions` (no tx exists today): wrap its
read-upsert in `InTx` (behavior-preserving: same single write);
`Old`/`New` populate only `PlanModeInstructions`; the two system-prompt
fields stay zero on both sides; no cross-key reads.
- Change detection compares the populated payload fields only (never the
artificial ID).
- Tests: handler-level coderdtest with `audit.NewMock()` asserting Write
entry on change and NO entry on a value-identical PUT, for both
endpoints (this also exercises `ResourceRequiresOrgID` end to end); the
fallback-flip case (no explicit include-default row, nonempty prompt set
to empty, effective boolean flips: entry emitted with the boolean diff);
diff assertions (old->new prompt text tracked, not secret) in
`enterprise/audit/diff_internal_test.go`; `TestAuditableResources`
passes by construction.
- Bookkeeping at PR open: correct CODAGT-719's no-op premise ("matches
the existing 204-on-unchanged behavior" does not exist on main;
suppression is new, write path unchanged).
- Review focus: Old capture and the New re-read inside the tx (three of
four existing singletons never set Old; do not copy them; and the
computed include-default value makes a naive New construction wrong);
the skip-on-no-op mechanism; prompt text deliberately visible in diffs.

</details>

Note: the plan excerpt above predates operator decision D5 (2026-07-30),
which this PR implements: the resource type is
`chat_instruction_settings` (not `chat_system_prompt_settings`), each
setting carries a stable ID and a display-name target (not a per-write
artificial ID and an empty target), no-op suppression runs through
`InitRequestWithCancel` (not the nil-ID skip), and attempts (denied,
failed, capture-degraded) record rows with real statuses and empty
diffs. Ticket bookkeeping for CODAGT-719 was corrected on Linear at
kickoff: the ticket's "matches the existing 204-on-unchanged behavior"
premise does not exist on main; suppression is new, and the write path
is unchanged.

> 🤖 This PR was created with the help of Coder Agents, and _will be_
reviewed by a human. 🏂🏻

---------

Co-authored-by: Michael Suchacz <203725896+ibetitsmike@users.noreply.github.com>
2026-08-18 20:03:06 +02:00
Michael Suchacz affeeaf9c8 feat: extend agent chat MCP tools for remote UAT evidence loops (#28233)
Extends the Agent-chat MCP tools so an unattended UAT evidence loop can
fetch artifacts, monitor long runs, and find prior runs without burning
model context.

## Backend

- New `chat_files_token` crypto key feature (migration 000571) with
rotator support and a dedicated signing keycache on coderd.
- `POST /api/experimental/chats/files/{file}/download-url`
(authenticated) mints a short-lived (5 min) signed URL and returns it
with `sha256`, `size_bytes`, `name`, `mime_type`, and `expires_at`.
- `GET /api/experimental/chats/files/{file}/download?token=` (no session
token) redeems the signed URL: verifies the JWS, requires the token's
`file_id` to match the path, and re-checks the minting user's RBAC
access live at redemption. Clients can `curl -o` artifacts with zero
credentials in the URL consumer.
- `ChatFileMetadata` gains `size_bytes` (via `octet_length`, no bytes
fetched).

## MCP tools (`codersdk/toolsdk`)

- `coder_download_chat_file`: by `file_id` or `chat_id`+`file_name`;
returns the signed URL plus checksum and size instead of base64.
- `coder_await_chat`: blocks (bounded `wait_secs`, 1-120) until a chat
leaves `running`/`interrupting`, using the existing watch stream with
subscribe-before-read.
- `coder_list_chats`: label, query, and limit filtering; chat
projections now include labels.
- `coder_get_chat_messages`: `after_id` forward cursor with
`next_after_id` (exact incremental reads), plus per-message `files`
metadata so artifact-bearing messages are identifiable.
- `coder_get_chat`: file listings now include `size_bytes` and
`created_at`.
- `coder_list_templates`: exposes `agents_allowed` for pre-flight
checks.

## Testing

- coderd: mint/redeem happy path with an unauthenticated client,
expired/tampered/file-mismatched tokens, auth still required on the
plain file endpoint, non-owner mint rejection.
- toolsdk: harness + integration coverage for all new/changed tools,
including signed-URL redemption with checksum verification,
forward-cursor exactness, await transition/timeout paths, and label
filtering.
- Remote dogfood UAT (dev.coder.com Coder Agent) passed all six
acceptance scenarios end to end over both MCP transports.

Note: `go test ./codersdk/toolsdk/` has a pre-existing goleak flake on
main (leaked `agentssh` non-PTY session goroutines from SSH exec tests;
reproduced 3/3 on clean `b4971bc49f1`). It is unrelated to this diff.

> Mux acted on Mike's behalf to create this PR.

<!-- mux-attribution: model=claude-sonnet-4-6 thinking=high -->
2026-08-18 19:15:30 +02:00
Michael Suchacz 7724ee281a feat: defer MCP tool schemas behind a find_tools search (#28225)
## Summary

When the `mcp-tool-search` experiment is enabled, chatd stops inlining
connected MCP tool schemas into every generation. It instead exposes a
built-in `find_tools` tool whose description carries a compact catalog
of the deferred tools, and only ships full JSON schemas for tools the
model has activated by searching or by calling them directly.

Closes [CODAGT-760](https://linear.app/coder/issue/CODAGT-760).

## Problem

Tool-heavy agent configurations (GitHub, Linear, Notion, and dev-tooling
MCP servers) inline over 100k tokens of tool schema definitions into
every generation. Initial uncached requests reached ~216k tokens with
time-to-first-token close to nine minutes, while the model typically
invokes only a handful of tools per turn.

## How it works

- `decideMCPToolSearch` defers external and workspace `.mcp.json` MCP
tools whenever the experiment is enabled. Native, dynamic, provider,
skill, and transport tools are never deferred.
- `find_tools` embeds a server-grouped catalog in its tool description
(degrading to names-only, then counts-only, then a constant-size summary
past a context-scaled size cap) and scores keyword matches across tool
names, descriptions, parameter schemas, and server metadata. Queries can
scope to one server with a `server:` prefix, and exact `names` arguments
always activate.
- Activation state is ephemeral: it is re-derived each generation from
surviving chat history (`find_tools` results and direct calls to
deferred tools), so activations naturally lapse when compaction
summarizes them away. Aggregate activated schema weight is capped at 10%
of the context window, shedding the least recently activated schemas
first; `find_tools` shares that budget across parallel calls in one
step. No new persistence.
- Deferred tools stay registered for execution, so the model can call a
cataloged tool directly without searching first; the schema is activated
for subsequent steps.
- Fail-open: the experiment being disabled, an empty candidate set, or
an MCP tool named `find_tools` all disable deferral, leaving today's
behavior byte-identical on the wire.
- Prometheus counters/histograms track `find_tools` calls, matches,
activations, and deferred token weight.
- The conversation timeline renders `find_tools` calls with a collapsed
search summary and expandable match list, falling back to the generic
renderer on malformed payloads.

## Validation

- Unit tests for the catalog, matcher, experiment-gated decision, and
activation derivation; end-to-end chatd generation tests covering
search-then-call, direct-call activation, experiment-off wire parity,
compaction lapse, and subagent tool gating.
- Storybook interaction tests for the timeline rendering and
malformed-payload fallback.
- Remote dogfood UAT on dev.coder.com passed: deferral with a real MCP
server and Anthropic model, direct calls without prior search,
activation persistence across turns, experiment-off parity, and clean
UI/console.

> Disclosure: Mux (AI agent) authored this PR on Mike's behalf.
2026-08-18 19:12:47 +02:00
Michael Suchacz 119f2b1dd9 feat: limit concurrent chat agents with pooled admission (#27902)
Limits concurrent chat generation on capped deployments to 5 root chats
and 10 delegated subagent chats. The pools are deployment-wide and
independent, so delegated work can continue while root capacity is full.

The default caps live in AGPL code. Enterprise contributes only a
licensing unlock, so unlicensed deployments stay capped and cannot fail
open. Licensed deployments are uncapped while Agent Hours usage stays
below an explicit hard limit. Deployments without a hard limit remain
uncapped, and reaching the Agent Hours allocation only triggers
warnings.

Admission happens before a worker takes chat ownership. Capped
deployments serialize admission across replicas with a
transaction-scoped advisory lock and derive active and queued state from
current ownership plus fresh runner heartbeats, rather than persisted
queue markers or per-replica state. The acquisition query returns a
bounded, pool-interleaved candidate set instead of ranking the whole
backlog; a migration replaces the acquisition index with a pool-aware
one. Refused chats stay running but unowned, and interrupt requests
bypass admission so users can stop queued or over-cap chats.

The single-chat API derives `queued_for_capacity` from live pool state;
list endpoints do not report it. The UI polls that value every 5 seconds
while a chat is running and shows a callout when the chat is waiting for
capacity.

Updates the administrator documentation and deployment-wide Prometheus
gauges for active and queued agents. Replica-level values must be
aggregated with `max`, not `sum`.

> Mux updated this PR on Mike's behalf.
2026-08-18 16:55:43 +02:00
Jaayden Halko fa8ffe4eda feat: report agent runtime hours usage in entitlements (#27985)
Populate `FeatureAgentRuntimeHours.Actual` on every entitlements refresh
for licenses that grant the feature. A new
`GetTotalUsageHBAgentRuntimeV1` query sums `runtime_ms` over the
license's usage period, reading `usage_events` directly:
`hb_agent_runtime_v1` is exactly one row per hourly bucket
deployment-wide with `created_at` at the bucket start, enforced by the
unique partial index introduced in #27983.

The measurement reuses the shared `measureUsage` policy from #27984
through a new `AgentRuntimeMsFn` closure (usage publisher subject):
failures publish the stable
`LicenseAgentRuntimeUsageUnavailableErrorText` and log the cause. Usage
is floored to whole hours, matching the unit of the
`agent_runtime_hours_*` claims, and at most one warning is emitted per
refresh: reaching the allocation supersedes the advisory soft limit. The
dashboard renders the soft-limit advisory muted without a sales link and
treats the runtime usage-unavailable text as a diagnostic.

**Precise usage.** `Feature.ActualMs` (JSON `actual_ms`), set only for
`agent_runtime_hours`, carries the exact stored milliseconds backing the
floored `Actual` so clients can render fractional hours (e.g. `10.3`).
It has the same freshness as `Actual`; the whole-hour warning thresholds
are unchanged.

**Unlimited licenses.** A license minted with the unlimited (`-1`)
allocation decodes to an enabled feature with a nil `Limit` (#27984), so
the warning write-back now guards the allocation dereference: no
thresholds can exist for an unlimited license, so no runtime hours
warning is ever emitted, while `Actual` is still measured and published.
`Feature.Compare` is unchanged; for usage-period features the
issued-at/end dates decide first, so a metered feature outranks an
unlimited one only on an exact timestamp tie, an edge pinned by a
`TestFeatureComparison` case and documented on
`decodeAgentRuntimeHours`.

**Grandfathered premium licenses.** Premium licenses without
`agent_runtime_hours_*` claims are now granted the feature disabled with
a zero limit over the license term, identical to an explicit
`allocation: 0`: usage is measured and published for every Premium
deployment, and chatd's pooled admission (#27902) caps concurrent
agentic chats until a license with a positive allocation is added. The
default carries a fixed early `UsagePeriod.IssuedAt` (2026-08-01, the
same mechanism as the managed-agents default) so any license actually
carrying the claims outranks it in the `AddFeature` merge regardless of
the licenses' relative issue dates; the constant must stay earlier than
the earliest legitimately issued claim-bearing license. Zero allocations
(explicit or grandfathered) emit no deployment-wide warning banner:
those deployments are steered by the in-page upgrade CTA and the
concurrency cap. Enterprise licenses are unchanged.

Part 3 of a 3-PR stack splitting up #27796 (see there for review
history). Stack: #27983 → #27984 → this PR.

Closes CODAGT-852.
2026-08-18 12:40:33 +07:00
Asher b5d18bb9c9 feat: add redirect URL override for external auth (#28082) 2026-08-17 14:09:23 -08:00
Susana Ferreira 2d9b6eda8f feat: add experimental CLI to price unpriced AI models (#27926)
## Description

AI Gateway computes the cost of an interception from `ai_model_prices`,
which is seeded on every server start from a price book embedded in the
binary. A model the price book does not cover records a NULL cost, so
its spend is invisible to cost reporting and is not enforced against
budgets. The only fix was to wait for a Coder release that added the
model.

This adds an experimental CLI, backed by an experimental HTTP endpoint,
for pricing those models. Models the price book already covers are
rejected, because the seeder re-applies the book on every start and
would overwrite an operator price. Support for custom pricing will be
handled in
https://linear.app/codercom/issue/AIGOV-589/extend-experimental-cli-command-to-set-custom-ai-model-prices.

## Commands

```
coder exp ai-model-prices list [--provider] [--model]
coder exp ai-model-prices update [file|-] [--provider] [--model] [--input-price] [--output-price] [--cache-read-price] [--cache-write-price] [--yes]
```

## Changes

- Add `GET` and `POST /api/experimental/ai/model-prices`, gated behind
the AI Bridge entitlement and the existing `ai_model_price` RBAC
resource.
- Add a `GetAIModelPrices` query with optional `provider` and `model`
filters applied in SQL.
- Validate the whole request before writing anything, so one bad entry
cannot leave the table half updated, and report every problem at once.
- Reject prices for models the embedded price book already covers,
through a new `prices.IsDefaultPriced`.
- Add the `coder exp ai-model-prices` command with `list` and `update`.
`update` accepts a JSON document or the single-model flags and prints a
plan, asking to confirm unless the document is piped in or `--yes` is
passed.
- Consolidate the supported provider list into
`coderd/aibridge/prices/providers` so the price generator and the server
share one definition.
- Add `codersdk` types and client methods for both endpoints, and bound
the request body at 1 MiB.
- Document the command in the AI Gateway cost controls page.

Closes
https://linear.app/codercom/issue/AIGOV-567/experimental-cli-command-to-set-prices-for-unpriced-ai-models

> [!NOTE]
> Initially generated by Claude Opus 5, modified and reviewed by
@ssncferreira
2026-08-13 15:00:36 +01:00
Nick Vigilante 0a7bb80a1e feat: make CLI/API doc generators emit front-matter metadata (Phase 2) (#27246)
## Summary

Phase 2 of the H1 → front-matter migration (`DOCS-483`; parent
`DOCS-477`). Makes the two reference-doc generators emit per-page
metadata as YAML front matter instead of a leading `# H1`, so generated
pages are self-describing and `make gen` stops reverting migrated pages
(Phase 3).

Phase 1 (`DOCS-482`) made the coder.com renderers prefer a front-matter
`title` (manifest fallback).

> [!NOTE]
> Rebased onto `main` and fully regenerated, and updated across two
rounds of Coder Agents Review — see **Review follow-ups** below.

## Changes

- **`scripts/clidocgen/command.tpl` + `gen.go` + `main.go`** — front
matter now carries `title` (from `fullName`) and `description` (from the
command's `Short`), and the leading `# H1` is dropped. The CLI index
page's front matter is taken from the manifest `Command Line` route
(title/description/icon_path).
- **`scripts/apidocgen/postprocess/main.go`** — reads the manifest and,
at write time, injects front matter carrying each section's `title` plus
any curated `description`, `state`, and `icon_path`. The API index
page's front matter is taken from the manifest `REST API` route.
- **`scripts/docgenenv`** (new shared code) — one `YAMLScalar`
front-matter escaper, one `Route`/`Manifest` schema +
`LoadManifest`/`FindRoute`, and one `FrontMatter(Route)` emitter, all
imported by both generators (no duplicated helpers, types, or emitters).
- Regenerated all **166 CLI + 31 API** reference pages.

### Metadata → front matter, and what stays in the manifest

Every *per-page* manifest field is mirrored into the page's front
matter: `title`, `description`, `state`, `icon_path`. The **structural**
fields stay in `manifest.json`:

- `children` — the nav tree (explicitly out of scope).
- `path` — the manifest's pointer to the file; a page carrying its own
path is redundant/error-prone, so it's treated like `children`.

The fields are **duplicated** into front matter and **`manifest.json` is
left unchanged**, so this is a **no-op for rendering today** (coder.com
strips front matter for `llms`, and Algolia + the renderer read only
`title`). Removing the fields from the manifest is the natural
follow-up, gated on the renderer reading them from front matter first.

### Why the API side changes the postprocessor, not the `.dot` templates

The issue text suggested editing
`scripts/apidocgen/markdown-template/*`. I deliberately did **not**,
because the postprocessor derives each page's **filename, section title,
and manifest route** from the leading `# {name}` line
(`extractSectionName`). Emitting front matter from the template would
break that extraction. Instead the widdershins templates still emit `#
{name}`, the postprocessor reads it (and now verifies it), and then
swaps the heading for a front-matter block as each section is written.

## Review follow-ups (Coder Agents Review)

### Round 1 — addressed in `e53d5e03` (all threads resolved)

- **CRF-1 / CRF-4** — de-duplicated the escaper and the
`route`/`manifest` schema + traversal into `scripts/docgenenv` (shared
by both generators).
- **CRF-2** — `YAMLScalar` now quotes YAML-reserved scalars
(`true/false/null/…`, numbers); no current value is affected.
- **CRF-3** — added unit tests: a `YAMLScalar` round-trip, `FindRoute`,
and `prependFrontMatter`.
- **CRF-5** — the CLI and API **index** pages now mirror their manifest
route's title/description/icon_path instead of a hardcoded
`coder`/`API`, fixing a rendered-heading regression (`REST API`/`Command
Line` were being overwritten).
- **CRF-6** — dropped the dead `#login` anchor in
`docs/support/support-bundle.md` (the migrated `login.md` no longer
mints that heading anchor).
- **CRF-7 / CRF-8 / CRF-11** — renamed to `prependFrontMatter`, switched
to `bytes.Cut`, and it now strips the first line only when it is the `#
{name}` heading (`extractSectionName` errors otherwise).
- **CRF-9** — removed the orphan `docs/reference/api/chat.md` (not in
the manifest, not linked; the real page is `chats.md`).
- **CRF-10** — the metadata read and the manifest rewrite now share one
`FindRoute` traversal.
- **CRF-13** — moot under squash-merge; this branch is a single
scopeless commit.
- **CRF-15** — the pre-existing `sort.Slice`/`slices.IsSorted`
comparator is left as-is per the review (out of scope; safe today
because section names are unique).

### Round 2 — addressed in `ee796e7107` (all threads resolved)

- **CRF-16** (P1) — removed three em-dashes from new doc comments (the
only `make lint` failure on the prior head); the emdash gate is green.
- **CRF-17 / CRF-18** — unified front-matter emission into one shared
`docgenenv.FrontMatter(Route)`, used by the API postprocessor directly
and by `command.tpl` via a `frontMatter` template func. This retires the
hand-written template YAML and the
`indexTitle`/`indexDescription`/`indexIconPath` closures, so a new
front-matter field is wired in one place, and it gives the CLI index the
`state` arm it previously lacked. Verified byte-identical: a full CLI +
API regen produces zero page changes.
- **CRF-19** — CLI child sort switched to `slices.SortFunc` +
`cmp.Compare` (typed comparator).
- **CRF-20** — reworded the `prependFrontMatter` comment:
`extractSectionName`'s fail-fast is the load-bearing guard; the prefix
check is a defensive backstop.
- **CRF-21** — added `icon_path`/`state` coverage in `docgenenv`'s
`TestFrontMatter/AllFields` (the branch the index page relies on,
previously at 0%).
- **CRF-22** — `YAMLScalar` no longer emits a trailing-space value as a
bare scalar (YAML strips it on read, so it would not round-trip); added
test coverage.
- **CRF-24** — the shared emitter removed the duplicated `cliIndexRoute`
doc comment; the rationale now lives in one place.
- **CRF-23** (Phase 3, out of scope here) — noted: the API generator
wipes and regenerates `reference/api/` from the manifest, so removing
curated metadata from the manifest in Phase 3 needs another source first
(a generator that preserves existing front matter, or metadata carried
alongside the swagger annotations).
- **Process (Mafu-san)** — the verification set below now leads with
`make lint`, the mandatory CI gate that the earlier list omitted.

## Cross-repo dependency

**Resolved — this PR no longer has a hard merge-ordering gate** (CRF-14
was right; the earlier "must merge after #968" note was stale).

The coder.com surfaces that would otherwise leak raw front matter from
`coder/coder` `main` are already front-matter-aware on merged PRs:

- **coder.com#964** (`DOCS-554`, llms-full.txt corpus + Algolia) —
**merged**.
- **coder.com#974** (`DOCS-574`, the `.md` proxy twin + `llms.txt` index
titles) — **merged**.

coder.com#968 (`DOCS-577`) was re-scoped to only the renderer
route-metadata generalization; it's a no-op on today's corpus and its
own description confirms the "deploy before the generators" constraint
no longer applies (that was driven by the llms corpus, now in #964).
Worth a final confirmation that #964/#974 are **deployed** before merge,
but there's no branch/PR ordering blocker left.

## Verification & evidence

AI was the primary author of this PR (see disclosure below); per the [AI
Contribution
Guidelines](https://coder.com/docs/about/contributing/AI_CONTRIBUTING)
here is manual verification.

- `make lint` (golangci-lint + the emdash gate) passes; `go build` / `go
vet` / `go test` are clean for the generators + `scripts/docgenenv`;
`pnpm check-docs` passes.
- `swagger.json`, `docs.go`, and `manifest.json` are **unchanged** —
metadata is duplicated into front matter; command/section names and
routes did not move.
- The diff is purely additive front matter
(`title`/`description`/`state`/`icon_path`) + the leading H1 removal; no
body reflow. A full CLI + API regen produces **zero** page changes
beyond the two index pages.

<details>
<summary>Terminal evidence</summary>

CLI `description` from the command's `Short` (`YAMLScalar` quotes when
needed, e.g. a `Short` with a colon):

```md
---
title: server
description: Start a Coder server
---
```

API pages inherit curated manifest metadata (only Agents/Chats have any
today):

```md
---
title: Chats
description: "REST endpoints for Coder Agents Chats API (programmatic agent sessions)."
state:
  - early access
---
```

Diff scope + "no body changes" proof (uses an explicit `base..HEAD`
range, so it actually tests the claim):

```
$ git diff --shortstat origin/main
 210 files changed, 1447 insertions(+), 344 deletions(-)
# = 166 CLI + 31 API reference pages + generators + scripts/docgenenv
# swagger.json / docs.go / manifest.json: NOT modified

# Every removed line under docs/reference is a leading "# H1"; nothing else:
$ git diff origin/main..HEAD -- docs/reference/ | grep '^-' | grep -v '^---' | grep -v '^-# '
(empty)

$ pnpm check-docs
Summary: 0 error(s)
```

</details>

Linear: DOCS-483

> This PR was created with AI assistance (Coder Agents).
2026-08-11 15:22:55 +00:00
Andrew Aquino 6e07e2610f feat: add paginated API endpoint for groups (#27603)
backend-only changes from #27271; see that PR for summary of changes +
implementation details
2026-08-10 13:23:14 -07:00
Steven Masley 9a57dfa642 feat: include agent metadata in workspace list responses (#27934)
Closes #27933. Related: #27897 (single-agent GET).

Agent metadata is only readable via a per-agent watch stream, so reading
it across N workspaces costs N+1 requests. This adds a batch read to the
list endpoint:

```text
GET /api/v2/workspaces?q=param:"pool=demo" include_agent_metadata:task_status
```

- New `include_agent_metadata` search key, repeatable and key-scoped. It
expands the response, it does not filter workspaces.
- `GetWorkspaces` aggregates the requested keys as JSON behind a `CASE`:
without opt-in the response is unchanged and the subquery never runs.
Runs only for the returned page, inside the same authorized query.
- Agents in the response gain `metadata`
(`[]codersdk.WorkspaceAgentMetadata`, `omitempty`), mapped by the
`workspace_agent_id` each element carries. The collection script is
omitted; it can be long.
- `codersdk.WorkspaceFilter` gains `IncludeAgentMetadata []string`.
- No wildcard, no schema change, no migration.

---

Authored by Coder Agents on behalf of @Emyrk.
2026-08-10 08:13:32 -05:00
Ethan 0ac23e3ee1 feat: add per-template Coder Agents access control (#27285)
Relates to CODAGT-713

Depends on #27284

This makes the per-template `agents_allowed` field authoritative in the API and chatd. It adds optional create and metadata update fields with the intended default and omission semantics, supports `agents-allowed:` template search, includes the value in telemetry, and makes `list_templates`, `read_template`, and `create_workspace` read the template row directly. Existing-workspace retries remain idempotent, and blocked same-organisation templates return an actionable message.

The experimental `/template-allowlist` routes remain temporarily because the shipped AI Settings page still calls them, but they no longer control chatd enforcement. #27514 moves that page to per-template metadata, #27515 removes the legacy storage, routes, SDK types, and utility, #27517 adds the CLI flags, and #27518 updates the platform controls documentation for the per-template model, directly addressing CRF-5 and CRF-6. The stack is intended to merge as a unit.
2026-08-06 14:14:37 +10:00
Michael Suchacz 4b9880afa6 feat: add --chat-hook-allow-insecure to allow plain HTTP chat hook URLs (#27896)
Adds a hidden `--chat-hook-allow-insecure` /
`CODER_CHAT_HOOK_ALLOW_INSECURE` deployment option (default `false`)
that allows the chat lifecycle hook URL to use plain HTTP for any host.

The HTTPS requirement is enforced at two points, and the flag relaxes
both: `DeploymentValues.Validate()` rejects `http` hook URLs at startup,
and the hook dispatcher's `validateHookURL` allows `http` only for
loopback hosts. With the flag set, any-host `http` is accepted; the
host, fragment/userinfo, secret, and timeout checks are unchanged, and
non-http(s) schemes still fail. This removes the need for an HTTPS
reverse proxy when testing a hook consumer on a trusted network.

Following security review feedback, the flag description and docs state
that plain HTTP lets an on-path attacker forge hook responses (which
control agent execution), and `coder server` logs a startup warning
(with a redacted hook URL) when hooks run over plain HTTP.

Docs, generated API types, and the server config golden are updated
accordingly.

> Mux acted on Mike's behalf to create this PR.
2026-08-05 22:41:17 +02:00
Nick Vigilante 9dcb75cd56 chore: add docs inline-HTML linter and backtick generated placeholders (#27399)
## What

Adds CI enforcement that fails when docs Markdown contains invalid
inline HTML
the docs site silently drops or mangles, and fixes the remaining
generated-doc
placeholders at their source.

This is the tooling half of the docs-HTML audit. The hand-written fixes
it
guards landed in #27298 (kept small and separate so it reviewed fast);
this PR
carries everything that touches code, CI, or generated output.

## Changes

**Linter (`scripts/docshtmlcheck`), wired into `make lint` via
`lint/docs-html`.**
Markdown-aware: parses each file with goldmark and inspects only
raw-HTML nodes,
so angle brackets in fenced code blocks, inline code, HTML comments, and
`<https://…>` / `<user@host>` autolinks are ignored. Flags swallowed
placeholders (`<region>`), void-element end tags (`</br>`), unregistered
or
incorrectly capitalized component tags (`<Image>`), and unclosed
container tags (a
`<div class="tabs">` that leaks its wrapper). The one intentional
renderer
component, `<children>`, is allowed but still balance-checked.

**Generator-source placeholder fixes (regenerated via `make gen`).**
- `codersdk/chats.go`: backtick `<server>__` in the
`ChatContextTool.Name` doc
  comment (it becomes the Swagger description, so it was swallowed in
  `reference/api/{chats,schemas}.md`).
- `codersdk/deployment.go`: backtick `<region>` in the AWS Bedrock
region flag
help (swallowed in `reference/cli/server.md`); also updates `coder
server
  --help` output and the golden files.

**Temporary allowlist.** `docs/reference/cli/agent-firewall.md`'s
`<host>` /
`<glob>` come from the external `github.com/coder/boundary` CLI help
(still
`v0.10.0` on `main`), so they are suppressed on that one file. The
suppression
is self-clearing: if an allowlisted tag stops appearing on a scanned
file, the
linter reports `stale-allowlist-entry` and fails until the dead entry is
removed, so a dead entry cannot silently mask a later regression of that
tag on
that page. (An entry whose file is deleted outright is never rescanned,
but a
missing file yields no findings, so nothing hides behind it either.)

## Review feedback addressed

This tool + generator work was reviewed by Coder Agents Review while it
was
bundled into #27298. Addressed here:

- **P1:** tokenize each raw-HTML node as a whole instead of per source
line, so
a tag whose attributes wrap across lines is no longer torn in half. This
fixes
both the missed multi-line unclosed `<div>` (a leaked wrapper that
passed with
exit 0) and the spurious `stray-end-tag` on valid multi-line tags. Each
token
  maps back to its own source line.
- Normalize allowlist lookup/report paths to a canonical repo-relative
form, so
the escape hatch no longer silently misses under absolute / `./` paths.
- Route generated-page findings to the generator source.
- Add `<search>` to the allowed set; reword the unknown-element message
to note
  that a real element can be added to `allowedElements`.
- Self-clearing allowlist guard (above); rename `optionalEndTag(s)` and
`kindUnclosed(Tag)`; adopt `slices`/`maps` idioms; move the lint banner
to the
  Makefile recipe; stop aliasing the input slice in `filterAllowed`.
- New tests: multi-line tokenization (both classes), interleaved
nesting, a
  pinned line number, `collectMarkdown`, and the stale-allowlist guard.

### Round 2 (Coder Agents Review on this PR)

A second `/coder-agents-review` pass on this PR raised 16 findings;
addressed in
`fix(docshtmlcheck): catch self-closing containers and capitalized
tags`:

- **P2:** self-closing container tags (`<div class="tabs"/>`) were
ignored by
the HTML5 parser and leaked their wrapper like the open spelling; the
balance
  check now tracks self-closing tokens too (CRF-1).
- **P2:** a capitalized component tag whose lowercase name is a real
element
(`<Table>`, `<Section>`) slipped through on the `allowedElements`
lookup. The
tokenizer lowercases tag names, so the check now reads the raw token and
  reports any capitalized name as a component reference (CRF-2).
- Narrowed the `:` / `@` autolink skip to a real URI scheme or a dotted
`local@domain`, so `<region:id>` and `<user@host>` stay checked (CRF-3).
- Stale-allowlist findings now report against the linter source with no
line,
and count separately from invalid-HTML issues in the footer (CRF-7,
CRF-11).
- Comment / README / Makefile wording synced to the honest
capitalized-tag
  behavior; added the deleted-file allowlist caveat and a note that
`allowedElements` is hand-maintained against the renderer (CRF-14,
CRF-17,
  CRF-9).
- Internal cleanups (`pop` -> `matchEndTag`, extracted
`unclosedFinding`) and
new tests: self-closing, capitalized open/close, colon/at placeholders,
a
non-first-token line assertion, `isGeneratedDoc`, and the stale message
  (CRF-12, CRF-13, CRF-1/2/3/4/5/16).

Two findings resolved without a code change:

- **CRF-8** (also wire `lint/docs-html` into `lint-light`): declined.
  `lint-light` is the Go-free fast path; `lint/docs-html` needs the Go
toolchain, so it stays in the full `make lint`, which CI runs. Adding it
would
  pull Go into the light path for no coverage gain.
- **CRF-9** (`allowedElements` <-> renderer coupling): documented with a
maintenance note in the `allowedElements` comment and tracked in
DOCS-597 for
  a cross-repo sync/check decision.

Deferred (note, no current trigger): raw-text element interiors
(`<script>` / `<style>`) are not scanned for nested tags. No docs page
relies
on this today; noted for follow-up.

## Merge order

#27298 (the hand-written fixes this PR guards) has merged, and this
branch is
rebased on `main`, so `make lint/docs-html` now reports 0 findings and
the
`lint` check passes. The two PRs are independent (disjoint files, no
stacking).

## Verification

- `go test ./scripts/docshtmlcheck/`, `go vet`, `gofmt -l`,
`golangci-lint run`: clean.
- `make lint/docs-html` (branch rebased on `main`): 0 findings.

## Linear

- DOCS-584:
https://linear.app/codercom/issue/DOCS-584/add-ci-check-that-fails-on-invalid-inline-html-in-docs
- DOCS-551:
https://linear.app/codercom/issue/DOCS-551/backtick-placeholder-syntax-in-generated-reference-docs-cli-help
- DOCS-597 (follow-up, from CRF-9):
https://linear.app/codercom/issue/DOCS-597/track-docshtmlcheck-allowedelements-drift-vs-docs-renderer-component

> This PR was created with AI assistance (Coder Agents).
2026-08-05 14:45:45 -04:00
Steven Masley 52423eb87b feat: promote MinimumImplicitMember experiment to GA (#27472)
Promotes the `minimum-implicit-member` experiment to GA and removes it.

## What changes

- The `minimum-implicit-member` experiment constant, its
`RoleOptions.MinimumImplicitMember` toggle, and the global
`rbac.MinimumImplicitMember()` accessor are deleted. The minimal-member
behavior is now the only behavior: `organization-member` and
`organization-service-account` carry only the floor (read-self records,
notifications, and similar) and grant **no workspace permissions**.
Workspace access lives exclusively on the
`organization-workspace-access` role.
- The experiment gate on customizing `default_org_member_roles` (`PATCH
/organizations/{org}`) is removed; the built-in-roles-only validation
remains.
- The dashboard's Default Roles section and the implied-roles display on
the members page are no longer experiment-gated.
- Admin docs: new "Default member roles" section in
`docs/admin/users/organizations.md`, cross-linked from
`groups-roles.md`.

## Why this is safe for existing deployments

Migration `000516` (shipped earlier) backfilled
`default_org_member_roles` with `['organization-workspace-access']` on
every organization. Members therefore keep exactly the effective
permissions they had with the experiment off; the workspace elevation
flows through the default role instead of being baked into
`organization-member`.

**Rollback caveat:** rolling back past this release restores the bundled
elevation, silently re-granting workspace access to members of
organizations that cleared their default roles.

## Review

Deep-review R1 findings are addressed in `chore: address deep-review
findings` (copy fixes, read-only Default Roles for viewers, removable
overlapping explicit grants, RBAC prose restoration, test
de-tautologizing, docs). Point-by-point disposition is in the PR
comments.

---

Generated by Coder Agents on behalf of @Emyrk.
2026-08-03 15:56:56 -05:00
Michael Suchacz df1c0f9710 feat: show what a chat lifecycle hook changed (#27655)
## Stack Context

Follow-up fixes from live UAT of the merged chat lifecycle hooks stack
(#27430). Two PRs:

1. **This PR**: make hook effects visible and correctly attributed in
the transcript.
2. [`mike/chat-hooks-uat/dispatch-capacity`]: reserve dispatch capacity
so an admission burst can't fail running turns.

## Why?

UAT found three ways the transcript misrepresented what a lifecycle hook
did. All three are user-visible and share the same surface
(`chathooks/effects.go`, `codersdk.ChatMessagePart`, the conversation
timeline), so they're reviewed together.

**A prompt `input_override` silently discarded attachments.**
`ComposeUserPromptContent` replaced the entire submitted part list with
one text part, dropping `file` and `file-reference` parts along with
their `chat_file_links`. The user saw their attachments vanish with no
explanation. The override now replaces submitted *text* parts only and
preserves non-text parts in order. A consumer that wants to block
attachments uses `deny`, which is the documented mechanism for refusing
a submission.

**Every user-visible `system` row was labelled "Lifecycle hook".** The
timeline keyed the notice off `role === "system"`. That was correct only
by accident, because the hook `user_message` was the sole client-visible
system row. The backend now emits the notice as a typed `hook-notice`
part and the timeline renders on that, so a future system row can't be
mislabelled as a policy notice.

**Nothing marked a tool call the hook had rewritten.** A consumer could
replace tool input via `input_override` and the transcript showed the
rewritten input as if the model had produced it. `ChatMessagePart` gains
`hook_rewritten`, set from `preflight.Overrides` on the same path that
already carries `ToolCallCreatedAt`, and the tool row renders a
"Modified by policy" badge.

`ToolCall.PolicyProvider` renders the badge itself, at four wrap sites:
the `Tool` dispatch wrapper, the `ReadFilesTool` aggregate and its
per-file rows, and `ReadFileTimelineBlock` (grouped and single
`read_file` rows bypass `Tool`). Renderer props do not include the flag;
descendants consume it through the provider context.

The badge is emitted by the provider rather than by the shared header
because several renderer branches return early without one, including
the auth-required `execute` card, a completed `ask_user_question`, and
an empty question payload. Those branches would drop the attribution
with no type or runtime error, and the gap is not greppable: every
renderer file contains a header somewhere, only individual branches do
not. Emitting at the provider removes the possibility instead of
enumerating the cases.

A rewritten call is wrapped in a group labelled by its badge, so one
rewritten file inside a merged read is attributed on its own rather than
inheriting the group's badge. `HeaderButton` still appends the policy
wording to an explicit `ariaLabel`, since an explicit `aria-label`
replaces the name computed from descendants.

Provider-executed calls are excluded from attribution. Hooks never see
them, and duplicate tool-call ID rejection deliberately skips them, so a
reused ID would otherwise mark a provider-executed call as
policy-rewritten.

## Testing

Go: `coderd/x/chatd/...`, `coderd/x/agenthooks/...`, `codersdk/...`, and
`coderd -run 'Hook|Chat'`. Frontend: `tsc` plus every `AgentsPage`
story; the only failures are `MCP Tool Completed` and `Scroll To Bottom
Button Works With Inverse Scroll`, both of which fail on trunk.

A registry-wide story asserts every registered renderer shows the badge,
verified against three inverted toggles: removing the badge, hiding it
with `display:none`, and skipping the provider for one renderer (which
names that renderer). Storybook also covers the rewritten subagent
spawn, a completed empty question payload, a non-hook system message,
and a failed `read_file` guarding the accessible name.

> Mux opened this PR on Mike's behalf.
2026-08-03 18:27:39 +02:00